Home / Blog

Axipro Resource Hub

Latest Articles

OWASP published the 2026 edition of its Top 10 for LLM Applications on August 4, 2026, during Black Hat week, and eight of the ten entries changed position. One got renamed. The message behind the reshuffle is blunt: you won’t build a model that can’t be fooled, so build the application around it in a way that limits the damage when it is. That one idea explains almost every move in the new ranking, and it should change how your team thinks about shipping AI features. This guide walks through the 2026 list in plain English: what each risk means, a real-world example, and what your team can actually do about it, with or without a dedicated security function. What Is the OWASP GenAI LLM Top 10 2026? The OWASP Top 10 for LLM Applications is a community-built awareness document that ranks the ten most critical security risks in applications powered by large language models. The OWASP GenAI Security Project, a global open-source initiative under the OWASP Foundation, maintains it, and the 2026 edition is the third release since the list first appeared in 2023. OWASP, the Open Worldwide Application Security Project, has published risk lists for web applications since 2003, and those lists became the shared vocabulary security teams, auditors, and buyers use to talk about risk. The GenAI LLM Top 10 does the same job for AI. Whether you’re a two-person startup wiring an API into a chatbot or an enterprise running retrieval pipelines, it gives you a common map of what actually goes wrong. One scoping note matters before anything else. The 2026 edition covers the model as a component inside an application: something that accepts input, generates output, and maybe retrieves information. The moment the model becomes an actor, with tools it can call and consequences it sets in motion, the risk shifts to the companion OWASP Top 10 for Agentic Applications from December 2025. Most products now do both, so most teams need both lists. Why the 2026 Update Matters for AI Builders Two things separate this edition from everything OWASP has published on AI so far. First, the methodology changed. Every previous version rested purely on expert consensus, meaning hundreds of practitioners voting on which risks matter most. This time the vote carried 75% of the weight, and the remaining 25% came from analysis of 6,639 real-world AI security incidents pulled from public vulnerability databases and an AI-harm database. It’s the first edition grounded in evidence of what has actually gone wrong rather than expert prediction of what might. Second, the framing changed. The project leads open the 2026 release by telling teams to stop optimizing the model and start optimizing the containment. The industry has spent two years pouring effort into filters, guardrail models, and jailbreak resistance. The 2026 list says: assume those will eventually fail, and make sure that when they do, nothing important breaks. AI security becomes blast radius control rather than perfect prevention. And this isn’t just a security engineer’s document. Developers decide what tools and permissions a model gets. Product owners decide which workflows run without a human in the loop. Founders and ops leads are the ones answering the security questionnaires where these questions now show up. The 2026 edition also ships a mapping appendix that connects every risk to frameworks your customers and auditors already recognize: NIST’s AI Risk Management Framework, MITRE ATLAS, MITRE CWE, and the Agentic Top 10. Insider Note: Enterprise vendor assessments have started asking about the OWASP LLM Top 10 by name. In security questionnaires we complete for clients at Axipro, questions like “describe your controls against prompt injection and excessive agency” began appearing in early 2026, sometimes before the buyer’s own team could explain what they meant. Being able to answer with a mapped control set is becoming a deal-cycle advantage, not just a security exercise. How the 2026 List Differs From Previous Versions The top two entries held their positions. Everything below them moved. Key Shifts Since the 2025 Update Excessive Agency jumped from sixth to third, the biggest promotion on the list. In 2025, giving a model tools and autonomy was mostly a theoretical worry. By 2026, agentic deployments had produced real production incidents, and the community concluded that agency is what decides whether a successful prompt injection is an inconvenience or a breach. Unbounded Consumption rose four places, from tenth to sixth. Inference costs became a real budget line as reasoning models, long outputs, and agent loops multiplied the compute behind a single request. “Denial of Wallet,” where an attacker spends pennies to trigger spend you can’t afford, is now a mainstream finding. Improper Output Handling fell from fifth to tenth. The risk didn’t shrink. It fell because it’s well understood and directly fixable with encoding and validation practices web developers already have. The entries above it are neither. What’s New, Renamed, or Reprioritized System Prompt Leakage became Hidden Context Exposure, and the scope widened a lot. The 2025 entry worried about attackers extracting your system prompt. The 2026 entry covers everything assembled into the model’s context that users aren’t meant to see: system instructions, retrieved policy documents, tool schemas, workflow rules. The guidance is unusually honest for a security document: assume all of it is discoverable, and design so that disclosure costs you nothing. Data and Model Poisoning absorbed fine-tuning subversion. The attack surface for corrupting a model’s behavior runs from pretraining data through fine-tuning pipelines into the retrieval stores RAG systems depend on, and the entry now says so. Misinformation climbed on evidence, not opinion. Practitioners voted it low; the incident data ranked it high. As reported in Help Net Security’s coverage of the release, OWASP also describes a “defense effect” working in the opposite direction on prompt injection: teams block it so effectively that few successful attacks reach public databases, which makes the risk look smaller than the money spent containing it. Signals About Where AI Security Is Heading Read together, the moves point one

For the past two years, enterprise AI risk conversations have centered on a familiar set of concerns: model bias, hallucination, data privacy, and dependency on third-party models. These are real risks, and most organizations now run some version of a governance program to manage them. But something has shifted. Organizations are no longer just deploying AI that generates content for a human to review. They’re deploying AI that acts. Agents now plan multi-step tasks, call APIs, move data between systems, execute transactions, and coordinate with other agents, often with no human checkpoint in the loop. That shift deserves more than a footnote in the existing AI risk category. It deserves its own line in the risk register: Agentic Autonomy Risk. What Is Agentic AI Risk Management? Agentic AI risk management is the practice of identifying, assessing, and controlling the risks created when AI systems take autonomous action on an organization’s behalf. Where traditional AI governance evaluates outputs (accuracy, bias, privacy), agentic AI risk management governs what agents actually do: the tools they call, the permissions they inherit, and the downstream consequences of their actions. That distinction is the reason existing risk registers struggle with agents, and it’s worth unpacking properly. What Agentic AI Actually Changes Traditional AI systems, even generative ones, are advisory. They produce an output such as a summary, a prediction, a draft email, or a classification, and a human remains the last checkpoint before anything happens in the real world. Agentic AI removes that checkpoint. An agentic system doesn’t just produce an answer. It pursues a goal. It decides which tools to call and in what order, then executes those actions directly against live systems: submitting a purchase order, modifying a database record, sending an external communication, or orchestrating a set of sub-agents to complete a broader workflow. Agentic autonomy is the degree to which a system can plan and execute actions without a human explicitly authorizing each step. It’s a spectrum rather than a binary. At one end, the AI drafts and a human approves every action. At the other, the AI operates within broad guardrails and only escalates exceptions. The further an organization moves along that spectrum, the less its exposure looks like software risk and the more it looks like delegated authority risk, the kind normally reserved for employees, contractors, and automated financial systems. Why Existing Risk Registers Miss Agentic AI Risks Most enterprise risk registers were built on a reasonably safe assumption: a human initiates consequential actions, and the technology around that human behaves deterministically. Agentic AI breaks both halves of that assumption at once. A few specific gaps show up quickly when organizations try to map agentic deployments onto existing categories. Operational risk registers assume process failures come from human error or system outages, not from a system independently choosing an unanticipated path to a stated goal. Cybersecurity risk registers are built around unauthorized external access, while an agent problem usually involves an authorized system taking unauthorized internal actions with its own legitimate credentials. Model risk frameworks, borrowed largely from financial services, evaluate output accuracy rather than action consequences, which matters most when those actions can’t be reversed. And third-party risk assessments treat vendors as static entities, not as autonomous agents that might invoke other vendors’ agents on your behalf. See our guide to the NIST AI Risk Management Framework for how output-focused frameworks are structured. The result is a governance blind spot. An organization can be compliant against its AI policy, its cybersecurity policy, and its vendor risk policy, and still have nobody accountable for the specific risk of a system initiating a harmful sequence of actions before anyone notices. Defining Agentic Autonomy Risk Agentic Autonomy Risk is the risk that an AI system, operating with delegated decision-making and execution authority, takes actions that are harmful, non-compliant, or misaligned with organizational intent before adequate human oversight can intervene. Those actions might happen independently or in coordination with other agents. It deserves standing as a named category alongside cybersecurity, operational, legal, financial, and third-party risk because the loss event itself is different. The harm is a completed action in a live system, and it may be difficult or impossible to reverse. The accountability structure is different too: when an orchestrating agent delegates to sub-agents, responsibility for the outcome gets distributed in ways existing ownership models don’t cleanly capture. So is the detection window. Traditional controls assume a human is positioned to catch an error before it compounds, but an agent can execute dozens of dependent actions faster than any human review cycle. 7 Agentic AI Risk Scenarios to Put on Your Register 1. Unauthorized autonomous decision-making. An agent takes an action within its technical permissions but outside its intended business mandate. It adjusts pricing, approves a refund, or modifies a customer record, and no policy ever explicitly authorized that scenario. 2. Goal misalignment. The agent optimizes for a literal interpretation of its objective in a way that diverges from actual business intent, particularly under ambiguous or adversarial inputs. 3. Multi-agent interactions and cascading failures. One agent’s flawed output becomes another agent’s trusted input. A single error can propagate across a chain of agents faster than anyone can detect it, amplifying the original mistake instead of containing it. 4. Excessive tool or system permissions. Agents get provisioned with broad, standing access “to be safe” rather than scoped, least-privilege access tied to specific tasks. A productivity tool quietly becomes a privilege-escalation path. 5. Regulatory non-compliance. Autonomous actions trigger obligations under data protection, financial services, employment, or sector-specific regulation, and they execute without the compliance review a human-initiated process would normally receive. 6. Explainability and accountability gaps. An autonomous action causes harm and the organization can’t clearly reconstruct why the agent chose that path, or establish whether the business owner, the AI governance function, or the vendor is accountable for the outcome. 7. Autonomous third-party actions. A vendor’s agent, integrated into your environment, takes action on your behalf, or your agent acts against a

A SOC 2 penetration test costs between $1,000 and $30,000 for most companies. A typical SaaS scope, meaning one web application, its API layer, and the cloud infrastructure behind it, usually lands between $2,000 and $20,000. Early-stage startups with a narrow scope can get an auditor-accepted test for $1,000 to $8,000, while enterprises with multiple products and hybrid infrastructure regularly spend $20,000 to $50,000 or more. The spread is wide because “penetration test” covers everything from an automated scan with a cover page to weeks of manual testing by senior engineers. Auditors know the difference, and so do the enterprise customers who asked for your SOC 2 report in the first place. This guide breaks down what drives the price, where the hidden costs sit, and how to buy a test that holds up in fieldwork without overpaying for it. What Is SOC 2 Penetration Testing?​ A SOC 2 penetration test is a simulated attack on your systems, performed by a qualified security professional, scoped to the environment covered by your SOC 2 report. The tester tries to exploit real weaknesses the way an attacker would: broken access controls, injection flaws, misconfigured cloud services, exposed credentials. The output is a report your auditor reads as evidence that your security controls work in practice, not only on paper. That last part matters. A pentest bought for SOC 2 has a second audience beyond your security team. If the report doesn’t map findings to your audit scope, document its methodology, and show remediation, it fails the job you bought it for. We cover the full deliverable in our guide to what a SOC 2-ready VAPT report includes. How Penetration Testing Fits Into SOC 2 Compliance​ SOC 2 is built on the AICPA’s Trust Services Criteria, and the Security category (the Common Criteria) applies to every report. Penetration testing is the standard way to satisfy CC7.1, which expects you to detect and monitor for new vulnerabilities, and it supports CC4.1, which covers ongoing evaluations of whether controls actually function. The AICPA’s points of focus explicitly mention vulnerability scanning and penetration testing as examples of how companies meet these criteria. In practice, the test slots into your audit timeline as an evidence item. Your auditor will ask for the report, check the test date against the audit period, and review how you handled the findings. Remediation is often scrutinized harder than the test itself, because it shows whether your vulnerability management process runs or merely exists. Is Penetration Testing Required for SOC 2?​ Strictly speaking, no. The Trust Services Criteria never use the word “mandatory” about penetration testing. You could theoretically satisfy CC7.1 with vulnerability scanning and strong monitoring alone. In reality, almost every auditor expects one, and skipping it invites two problems. First, your auditor may push back during fieldwork or add exceptions to the report. Second, the enterprise buyers reviewing your SOC 2 report increasingly look for pentest evidence specifically, and a report without it raises questions during procurement. Treat the test as effectively required and budget for it from the start of your SOC 2 compliance checklist. How Much Does SOC 2 Penetration Testing Cost? Typical Price Range for SOC 2 Pen Testing Most companies pay $1,000 to $30,000, with the median engagement for a SaaS business sitting around $12,000 to $15,000. Compliance-focused tests at the lower end of the market start around $1,000 to $5,000. Deep manual testing from established firms runs $10,000 to $30,000. Anything quoted below roughly $3,000 is almost certainly automated scanning packaged as a pentest, which auditors are getting better at spotting. Cost by Company Size (Startup, SMB, Enterprise) Company size is a proxy, not the driver. A 15-person company with three products and a legacy on-prem component will pay more than a 200-person company with one tightly scoped SaaS platform. Testers price effort, and effort follows scope. Cost by Test Type (Network, Web App, API, Cloud, Internal/External) Most SOC 2 engagements bundle two or three of these. The common package for a cloud-native SaaS company is web app plus API plus cloud configuration, which is why the $1,000 to $20,000 band comes up so often. Companies with office networks and internal systems in their audit scope add internal network testing, and the price climbs accordingly. Factors That Influence SOC 2 Penetration Testing Cost Scope and Number of Assets Tested Scope is the single biggest cost driver. Every additional application, API endpoint group, cloud account, or network segment adds testing hours. A pentest priced without a scoping call is a pentest priced on guesswork, and the guess usually favors the vendor. Complexity of Application or Infrastructure​ A simple CRUD app with two user roles tests quickly. A multi-tenant platform with role hierarchies, workflow engines, file processing, and third-party integrations takes far longer, because each of those features creates attack surface a tester has to work through manually. Authentication tiers matter especially: every distinct role needs testing for privilege escalation and cross-tenant data access. Testing Methodology (Black Box, Grey Box, White Box) Black box testing gives the tester nothing but a URL, grey box adds credentials and documentation, and white box adds source code and architecture diagrams. Grey box is the default for SOC 2 and usually the best value, since the tester spends time exploiting rather than discovering. White box costs more upfront but finds deeper issues. Black box sounds rigorous but often wastes paid hours on reconnaissance an attacker would run for free. Depth of Testing and Manual vs. Automated Approaches Automated scanning finds known vulnerability patterns. Manual testing finds business logic flaws, chained exploits, and authorization gaps that no scanner catches, and it’s the part auditors and security-literate customers actually value. The ratio of manual work to automation is the honest explanation for most price differences between two quotes covering the same scope. Tester Credentials and Firm Reputation Senior testers holding OSCP, GPEN, or CREST credentials bill higher rates, and firms with recognized methodologies charge a premium for the credibility their letterhead carries

Two compromised versions of LiteLLM sat on PyPI for roughly 40 minutes on the morning of March 24, 2026. That window was enough to capture secrets from around 434,000 CI/CD pipeline runs across nearly 2,500 organizations, including AWS, Samsung, Cisco, Salesforce, Siemens, and Deloitte. In August, researchers at CloudSEK and Hudson Rock confirmed they had obtained the raw exfiltrated data: a 153GB archive containing 433,909 files of environment variables, cloud keys, Kubernetes secrets, and API tokens harvested live from running pipelines, as covered by Help Net Security’s reporting on the credential archive. If LiteLLM runs anywhere in your stack, or you touch any AI proxy infrastructure at all, you need answers to three things: whether you were exposed, what to rotate first, and whether the rotation you did back in March actually held. That last one matters more than it sounds, because “we rotated everything” has already burned at least one very large company. How the Breach Happened The attack didn’t start with LiteLLM. On March 19, 2026, a threat group called TeamPCP compromised the build pipeline of Trivy, a vulnerability scanner half the industry runs, and pushed a poisoned release. LiteLLM’s own CI pipeline ran Trivy, so the poisoned scanner had legitimate read access to the project’s runner environment. The attackers used that to steal LiteLLM’s PyPI publishing tokens and ship two malicious releases of their own: versions 1.82.7 and 1.82.8. KICS and the Telnyx Python SDK got hit in the same campaign. The payload design is the part worth studying. The malicious package dropped a .pth startup hook into site-packages, so the code ran the moment any Python interpreter started on the machine, whether or not anything imported LiteLLM. From there it harvested environment variables, read local credential files like .aws/credentials and .kube/config, tried to move laterally across Kubernetes clusters, and installed a systemd backdoor dressed up as a generic telemetry service. InfoQ’s coverage of the PyPI compromise put downloads of the compromised release above 40,000. For scale, LiteLLM normally gets downloaded around 3 million times a day. The exfiltration had a nasty fallback, too. According to CloudSEK, stolen data was encrypted and sent to a typosquatted domain, and when that failed, the malware created a public repository inside the victim’s own GitHub account and uploaded the loot as a release asset. Some companies were publishing their own secrets to the open internet and had no idea. Worth Knowing: The malicious code only existed in the PyPI artifacts. The GitHub source repository stayed clean the whole time, so a developer reviewing the code on GitHub saw nothing wrong. Source review isn’t artifact verification. If you don’t check that what the registry serves matches the upstream source, this class of attack is invisible to you. How to Check If You Were Exposed Three checks, from quickest to most involved. 1. Confirm whether the compromised versions ever ran The malicious versions went live on PyPI at 10:39 UTC on March 24, 2026 and got quarantined about 40 minutes later. The project’s advice: treat any install from that day before 16:00 UTC as suspect. Search your lockfiles, pip caches, SBOMs, and container image histories for 1.82.7 and 1.82.8. And check your internal artifact mirrors. An Artifactory or Nexus proxy that cached the bad release in March can keep serving it internally long after PyPI pulled it. Keep the .pth mechanism in mind when you scope this. The question isn’t “which applications import LiteLLM,” it’s “which machines had the package installed at all,” because every Python process on an infected machine triggered the payload. 2. Hunt for persistence Rotation is pointless if the attacker still has a foothold. Check developer machines, CI runners, and containers for unauthorized .pth files in site-packages and for suspicious systemd units, especially anything posing as a system telemetry service. And review activity from March 24 onward, not just the 40-minute window. Persistence is there so the access outlives the infection. Pro Tip: Don’t limit the persistence hunt to live machines. Base container images rebuilt in late March may have baked the payload into every image derived from them since. Scan your image registry for the affected LiteLLM versions and for unexpected .pth files, then trace which running workloads came from flagged images. 3. Check whether your secrets are in the dump Hudson Rock has published a domain lookup tool and is running ethical disclosures for affected organizations, and CloudSEK maintains a high-confidence victim list. Use them, but know their limits. Attribution in this dataset is genuinely hard. One dump with a siriusxm.com committer email actually traced, through its self-hosted GitLab endpoints, to AdsWizz, a SiriusXM subsidiary. And a large share of the dumps are generic pipeline configurations with no identifying domain, email, or server name at all. Absence from a victim list is not evidence of absence. If your pipelines ran the compromised versions, assume exposure no matter what a lookup tool tells you. What to Rotate, in What Order The guidance from both research teams is blunt: treat every secret the LiteLLM environment could reach as compromised. That covers secrets on disk, in memory, injected into CI jobs, and anything retrievable through instance metadata services. Work down by blast radius: Priority Credential type Why it comes first 1 Cloud IAM keys (AWS, GCP, Azure) Direct control of infrastructure, data stores, and billing. This is where attackers monetize fastest. 2 GitHub and GitLab PATs, package publishing tokens These let an attacker poison your releases and turn your company into the next link in the supply chain. 3 Kubernetes service account tokens and kubeconfigs Lateral movement across clusters was built into the payload, not a theoretical risk. 4 Database passwords and third-party API keys Dumped in plain text in the archive, often with no attribution, so nobody will warn you they leaked. 5 AI provider API keys Billing abuse, quota theft, and access to whatever data flows through your LLM routing layer. One word matters more than the rest of this article: revoke, don’t just rotate. That

The EU AI Act names recruitment AI as high-risk. Annex III explicitly lists AI systems used for recruitment, candidate selection, and employment decisions, which pulls CV screeners, video interview platforms, and assessment tools into the most demanding compliance regime the Act contains. The original compliance date for these systems was August 2, 2026. In June 2026, the EU’s Digital Omnibus moved the deadline to December 2, 2027, a 16-month extension that has led many HR and talent teams to shelve the topic entirely. That’s a mistake, for two reasons. First, one rule that directly affects recruitment technology is already in force: the ban on emotion recognition in the workplace has applied since February 2, 2025, and it catches features still shipping in some video interview products today. Second, the deferred obligations didn’t shrink. Conformity assessments, human oversight design, bias monitoring, and documentation all still arrive in full, and the practical work of auditing a recruitment stack, renegotiating vendor contracts, and training hiring teams routinely takes a year or more. Here’s what the EU AI Act actually requires of employers and vendors using recruitment tools, on the timeline that now applies. Why Recruitment Tools Are Classified as High-Risk Under the EU AI Act​ Definition of High-Risk AI Systems in Hiring​ The Act takes a list-based approach. Annex III, point 4, designates as high-risk any AI system intended for the recruitment or selection of natural persons, including placing targeted job advertisements, analyzing and filtering applications, and evaluating candidates. The same point covers AI used for decisions on promotion, termination, task allocation, and monitoring of workers, so the classification follows the tool through the entire employment lifecycle, not just the hiring funnel. The reasoning is straightforward: hiring decisions shape access to livelihoods, and algorithmic discrimination in hiring is well documented. The European Commission’s regulatory framework for AI treats employment as one of the areas where an AI error or bias causes serious harm to fundamental rights. That’s the test for the high-risk tier. Types of Recruitment Tools Affected In practice, the high-risk classification captures most of the modern recruitment stack: CV and resume screeners that rank or filter applicants, video interview platforms that score responses or delivery, psychometric and skills assessment tools that produce scores feeding a hiring decision, sourcing and matching algorithms that decide which candidates a recruiter sees, and programmatic job ad targeting systems that determine who sees a vacancy at all. If the system’s output materially influences who advances and who does not, assume high-risk until proven otherwise. Important: Emotion recognition is not high-risk in the workplace. It is prohibited. Article 5 bans AI systems that infer emotions of people in the workplace (outside narrow medical and safety cases), and that ban has applied since February 2025 with the Act’s top penalty tier attached. If your video interview vendor markets “engagement scoring” or “sentiment analysis” of candidates, that feature needs to be switched off for EU hiring now, not in 2027. Recruitment Tools That May Fall Outside High-Risk Classification Not everything in the HR stack qualifies. The Act carves out systems performing narrow procedural tasks that do not materially influence decision outcomes. An applicant tracking system that stores applications, schedules interviews, and sends templated emails is a database with a workflow, not a high-risk AI system. The same goes for tools that transcribe interviews without scoring them, deduplicate candidate records, or generate first drafts of job descriptions for a human to edit. The line is decision influence: the moment a tool ranks, scores, filters, or recommends candidates, it crosses into Annex III territory. Deployers who rely on an exemption must be able to document that assessment, so “we decided it doesn’t count” needs to exist on paper. Extraterritorial Scope: Which Employers Are Covered The Act applies to providers placing AI systems on the EU market and to deployers established in the EU, but it also reaches further: it covers providers and deployers located outside the EU where the output of the system is used in the EU. For recruitment, the consequence is blunt. A US or UK company with no EU entity that uses an AI screener to filter applicants for roles based in Berlin or Dublin, or that screens candidates located in the EU, is using the system’s output in the Union. Brexit doesn’t move UK employers out of scope when they hire into or from the EU. Providers vs. Deployers of Recruitment AI Tools The Act splits obligations between the provider (the vendor that develops the tool and places it on the market) and the deployer (the employer using it). Most employers are deployers, and deployer obligations are lighter but real. One common trap: an employer that substantially modifies a high-risk system, or puts its own name on it, can be reclassified as a provider and inherit the full provider stack. Heavy customization of a screening model, or fine-tuning it on your own hiring data, can be enough to trigger this. Key Obligations for Employers Using AI Recruitment Tools Human Oversight in Automated Hiring Decisions Deployers must assign oversight of the system to people with the competence, training, and authority to intervene. That last word matters. A recruiter who rubber-stamps whatever the ranking algorithm produces, because nobody has time to review 800 rejected CVs, doesn’t count as oversight. Regulators and courts will look at whether the human could genuinely override the system and whether they ever did. Designing review checkpoints where a person can meaningfully change the outcome, and logging when they do, is the core of compliant deployment. Transparency Requirements Toward Candidates Employers must inform workers and their representatives before putting a high-risk AI system into use at work, and candidates subjected to such a system must be told it is being used. In countries with works councils, such as Germany, this obligation lands on top of existing co-determination rights, so employee representatives may need to be consulted before the tool goes live rather than just told afterward. Burying an AI disclosure in a privacy policy paragraph is unlikely to survive scrutiny.

A green dashboard is not an audit opinion. Compliance automation platforms like Vanta, Drata, Secureframe, and Hyperproof have made SOC 2 readiness faster and cheaper, but every audit cycle produces the same pattern: controls that sat at “passing” for months come back from the auditor with exceptions or requests for re-testing. The four controls below account for a disproportionate share of those rejections, and they all fail for the same underlying reason. The tool confirmed that evidence exists. The auditor tested whether the control actually operated. This article walks through each of the four: what auditors reject, why, and how to fix the evidence before fieldwork starts. Why Compliance Tools Show “Passing” But Auditors Still Reject Controls​ The Gap Between Automated Checks and Auditor Judgment Compliance platforms run continuous control monitoring: API calls that check whether a configuration exists, a document is uploaded, or a task is marked done. That’s real value. It catches drift, keeps evidence in one place, and saves weeks of screenshot collection. An audit is a different exercise. A SOC 2 examination is an attestation performed by a CPA firm under AICPA standards, and the auditor’s job is to form an independent opinion on whether your controls met the Trust Services Criteria. That opinion rests on professional judgment, not on whether an API integration returned a 200 response. What “Passing” Actually Means in Your Compliance Dashboard​ When a control shows “passing,” the platform is telling you one narrow thing: at the moment of the last scan, an automated test found the artifact or setting it was programmed to look for: MFA enforced in the identity provider, a policy document uploaded, a training campaign sitting at 100%. The test says nothing about whether the underlying process ran the way your control narrative claims it did, or whether it ran that way across the whole audit period. How Auditors Evaluate Controls Beyond the Checkbox Auditors test two dimensions. Design effectiveness asks whether the control, as described, would meet the criterion if it worked as intended. Operating effectiveness, the core of a SOC 2 Type 2 report, asks whether it actually did throughout the audit period. To answer that, the auditor pulls a population (every access review, every change, every new hire in the period), selects a sample, and inspects the evidence item by item. A dashboard status feeds into that process. It doesn’t replace it. Insider Note: Auditors increasingly ask for evidence outside the compliance platform precisely because they know what the platform auto-collects. If every artifact you produce comes from the same tool export, expect the auditor to independently pull the population from the source system and compare. Discrepancies between the two are one of the fastest routes to an exception. Control #1: Access Reviews That Automation Marks Complete but Auditors Reject Why Auditors Reject Automated Access Review Evidence​ User access reviews sit under the logical access criteria (CC6.1 through CC6.3), and they are the single most common source of audit exceptions we see. The typical failure: the platform generated a user list, someone clicked “complete,” and the dashboard turned green. The auditor then asks a simple question the evidence can’t answer: what did the reviewer actually decide? The Missing Element: Documented Reviewer Judgment​ An access review is a judgment control. Someone with knowledge of the system must look at each account and confirm the access is still appropriate for the person’s role. A timestamped task closure proves the task was closed. It doesn’t prove anyone assessed anything, and an “approve all” review completed in ninety seconds gets exactly the skepticism it deserves. What Auditors Actually Want to See in Access Review Evidence Auditors look for four things: The full population of accounts at the time of review (including service accounts and admin roles), Evidence of who reviewed it and when, explicit dispositions per account or group (retain, modify, revoke), and Proof that flagged access was actually removed. That last item, the deprovisioning ticket showing revocation within a defined window, is the piece most companies can’t produce. How to Fix Your Access Review Control Before the Audit​ Assign a named control owner per in-scope system, run reviews quarterly, and require reviewers to record a disposition for every line, not a blanket approval. When access is revoked, link the removal ticket to the review record. If a quarter was missed, don’t backfill it. Document it honestly and show the remediation, because auditors treat fabricated retroactive evidence far more severely than a disclosed gap. Control #2: Change Management Approvals That Pass Automated Scans​ Why Ticket Closure Isn’t Proof of Approval​ Change management (CC8.1) automation typically verifies that production changes link to a ticket and the ticket is closed. Auditors test something stricter: that each sampled change was approved by an authorized person before deployment. An approval added after the merge, or a ticket closed by the same engineer who wrote the code, fails that test even though every automated check came back green. The Segregation of Duties Problem Automation Misses Segregation of duties is the requirement that no single person can develop, approve, and deploy the same change. NIST’s SP 800-53 control catalog treats it as a foundational access control principle, and SOC 2 auditors apply the same logic. Small engineering teams trip on this constantly. Self-approved pull requests, admins who can bypass branch protection, direct pushes to main: a scanner sees “changes with tickets” while an auditor sees SoD violations. Emergency Changes and Retroactive Approvals: Common Rejection Triggers​ Every audit period contains hotfixes. Auditors don’t reject emergency changes. They reject emergency changes with no documented post-hoc review. If your policy says urgent changes get retroactive approval within two business days, the auditor will sample your emergency changes and check exactly that. No policy, or a policy nobody followed, produces an exception. Rebuilding Change Management Evidence Auditors Will Accept​ Enforce the control technically: branch protection requiring at least one independent reviewer, no admin bypass, and deploy pipelines that only run from protected branches. Then write the emergency change procedure down and generate the review artifact every time it fires.

AI Tool Usage Tracking

One in five breached organizations last year traced the incident to shadow AI, and those breaches cost an average of $670,000 more than standard incidents, according to IBM’s 2025 Cost of a Data Breach Report. The worst part is that most of those organizations already ran a CASB, a DLP program, or both. The tools were on, but the traffic still got through. That’s the visibility gap this article is about. AI tool usage tracking isn’t the same problem as SaaS discovery, and the security stack built for the SaaS era misses most of what matters about AI. Below, we break down what tracking actually requires, where CASB and DLP fail, which categories of AI usage slip through, and what a stack that works looks like in 2026. What AI Tool Usage Tracking Actually Means Most teams that say they “track AI usage” mean they can see that someone visited chat.openai.com. That is app discovery, and it answers almost none of the questions a security or governance team actually needs answered. Beyond App Discovery: Tracking Prompts, Data Flows, and Model Interactions Real tracking covers three layers. First, which tools are in use: chatbots, copilots, coding assistants, embedded SaaS features, agents. Second, what data moves: the content of prompts, uploaded files, and pasted context, mapped against data classifications. Third, how models behave in your environment: which endpoints get called, which OAuth grants exist, which agents hold standing permissions. Seeing that an employee opened ChatGPT gets you nowhere. What you actually need to know is whether they pasted a customer contract into a personal account while they were there. The Difference Between Detection, Monitoring, and Continuous Tracking Detection is a point-in-time answer to “what AI is here?” Monitoring watches known tools on an ongoing basis. Continuous tracking is broader: it assumes the inventory changes weekly, correlates identity, data, and endpoint signals over time, and feeds a governance program rather than a one-off report. Frameworks such as the NIST AI Risk Management Framework and ISO 42001 assume the third mode. A discovery scan from last quarter won’t satisfy an auditor, and it certainly won’t slow down an attacker. Why Traditional SaaS Monitoring Falls Short for AI SaaS monitoring was built around a stable premise: an app is a destination with a domain, a login, and an admin console. AI breaks that premise in several ways at once. The risky activity is the content of an interaction, not the visit. The tool often isn’t a destination at all but a feature inside an app you already sanctioned. And increasingly the “user” isn’t a person but an agent acting on delegated credentials. Why CASB Misses Shadow AI Usage The Cloud Access Security Broker sits between users and cloud services to enforce policy, and for classic SaaS governance it still earns its keep. AI has structural blind spots that no amount of tuning can fix. CASBs Were Built for SaaS Apps, Not Model Endpoints A CASB catalog maps domains to applications with risk scores. AI usage doesn’t resolve neatly to a domain. The same api.openai.com endpoint serves a sanctioned enterprise deployment, a developer’s weekend experiment, and a data-leaking browser extension, and the catalog sees one “app”. Meanwhile, new model endpoints, wrappers, and niche AI tools appear faster than any vendor catalog can keep up with. Gartner research from late 2025 found 69% of organizations already suspect or have evidence that employees use prohibited public generative AI tools, catalog or no catalog. Blind Spots in Encrypted API Traffic to LLM Providers Prompt content travels over TLS. Without full TLS inspection, a CASB sees connection metadata: destination, volume, timing. It can’t see that the payload contained source code or patient records. And full TLS inspection is harder than the datasheet implies. Certificate pinning breaks it for many native apps and CLI tools, legal and works-council constraints limit it in the EU, and most organizations carve out broad exemption lists that AI traffic happily rides through. The OAuth and Embedded AI Problem CASBs Can’t See When an employee grants an AI meeting-notes tool access to their calendar and mailbox via OAuth, no proxy is involved at all. The vendor’s servers communicate directly with Microsoft’s or Google’s APIs using a persistent token. The same applies to AI features embedded inside sanctioned SaaS, think Notion AI, Slack AI, or Salesforce Einstein. The CASB sees approved traffic to an approved app, while the AI processing happening inside it, and whichever sub-processor it forwards data to, stays invisible. Personal Accounts and BYO-AI Bypass CASB Proxies Netskope’s 2026 Cloud and Threat Report found that nearly half of employees who use generative AI at work do so through personal accounts. Personal accounts on managed devices are hard enough; personal accounts on personal devices, home networks, and mobile connections never touch the corporate proxy path at all. Tenant restrictions help for a handful of major providers and do nothing for the long tail. Browser-Based and Extension-Delivered AI Escape Network Inspection AI browser extensions read page content and form inputs locally, then exfiltrate via their own backend, often to generic cloud infrastructure that categorizes as “technology” rather than “AI”. From the network’s view, it is routine HTTPS to a CDN. The riskiest interaction, an extension scraping everything an employee views, produces the most boring traffic signature. Insider Note: In AI governance readiness assessments, the OAuth grant review is where clients get the biggest surprise. We routinely find dozens of AI tools holding live mail, calendar, or drive scopes that nobody in IT ever approved, granted by employees who abandoned the tool (and sometimes the company) months earlier. The tokens keep working anyway. Why DLP Fails to Catch Shadow AI Data Exposure DLP has the opposite problem. It can sometimes see content, but it doesn’t understand it, and AI interactions defeat the pattern matching it depends on. Prompt-Based Data Loss Doesn’t Match DLP Signature Patterns DLP fires on signatures: credit card regexes, SSN formats, keyword dictionaries, file fingerprints. Sensitive prompts rarely look like that. “Summarize why we’re losing

Most SOC 2 preparation effort goes into access controls, encryption, and vendor reviews. Then the auditor’s first evidence request arrives, and item one has nothing to do with technology: show us your board charter, your meeting minutes, and proof that your board operates independently from management. That’s CC1.2, and it causes more last-minute scrambling than almost any technical control in the framework. This guide explains what CC1.2 requires, provides a board charter template with sample language that auditors accept, and covers the situation most startups actually face: satisfying the criterion without a traditional board of directors. What Is a SOC 2 Board Charter and Why It Matters for CC1.2 A board charter is a formal document that defines your board’s purpose, composition, authority, meeting procedures, and oversight responsibilities. Outside of compliance, it’s a corporate governance tool and a good idea in general for companies with shareholders.  Inside a SOC 2 audit, it’s the primary design evidence for CC1.2, the criterion that asks whether an independent body oversees management and the internal control environment. The charter matters because CC1.2 is one of the few criteria where the control is a document plus behavior. The charter establishes the structure. The auditor then tests whether the structure operates: did the board actually meet, did it review the security program, did it challenge management? A beautifully drafted charter with no meeting minutes behind it fails just as surely as no charter at all. If you’re earlier in your preparation, our complete SOC 2 guide covers how the full audit fits together. Understanding CC1.2: The Board Independence Criterion CC1.2 is part of the Trust Services Criteria published by the AICPA (American Institute of Certified Public Accountants). The criterion requires that the board of directors, in the AICPA’s words, “demonstrates independence from management and exercises oversight” of how internal control is developed and how it performs. That sentence hides two separate tests. Independence means the board isn’t just management wearing a second hat. Active oversight means the board actually reviews and challenges the control environment instead of existing on paper. Plenty of companies pass one and fail the other. How CC1.2 Fits Within the CC1 Control Environment The Common Criteria run from CC1 through CC9, and the CC1 series covers the control environment: the governance and people layer everything else rests on. CC1.1 addresses integrity and ethical values, CC1.2 addresses board independence and oversight, CC1.3 covers organizational structure and reporting lines, CC1.4 covers competence and hiring, and CC1.5 covers accountability. CC1.2 is the layer that makes the other four credible. A code of conduct means little if nobody independent of management ever checks whether leadership follows it. The COSO Principle 2 Connection The Trust Services Criteria are built directly on the COSO Internal Control—Integrated Framework and its 17 principles. CC1.2 maps to COSO Principle 2, which carries four points of focus: the board establishes oversight responsibilities, applies relevant expertise, operates independently of management, and provides oversight of the system of internal control. Those four phrases are worth memorizing, because they’re effectively the outline of a good board charter. Why Auditors Prioritize Board Charter Evidence Auditors test the control environment first because failures there cascade. If governance is weak, every other control claim gets harder to trust: who approved the risk assessment, who reviewed the incident report, who held management accountable when a control slipped? An exception at CC1.2 tells the auditor that nobody independent was watching, and they’ll read the rest of your evidence with that in mind. That’s why board charter requests sit near the top of almost every evidence list. Worth Knowing: Points of Focus Points of focus are not pass/fail requirements. The AICPA describes them as characteristics that assist evaluation, and the 2022 revisions changed points of focus without changing any criteria. In practice, though, they function as the auditor’s mental checklist, so drafting your charter against them is the safest move. What Auditors Actually Look For in a Board Charter Auditors don’t grade prose style. They scan for specific, verifiable commitments. Here’s what they check, roughly in order. Documented Board Independence from Management The charter must state how many members are independent, define what independence means (no operational role, no material financial relationship beyond board compensation or equity), and describe how independence is maintained. “The board includes members independent of management” without a definition is boilerplate; auditors want criteria they can test against actual member profiles. Defined Oversight Responsibilities This is the heart of CC1.2. The charter should explicitly assign the board oversight of internal control, information security, and risk management. If the charter only mentions financial oversight and strategy, it wasn’t written with SOC 2 in mind, and the auditor will notice the gap. Clear Authority and Decision-Making Powers What can the board approve, veto, or demand? Typical provisions include approving the risk management framework, reviewing audit results, approving executive appointments, and requiring management to report on control deficiencies. Authority without teeth reads as decorative. Meeting Cadence and Quorum Requirements The charter should commit to a minimum meeting frequency (quarterly is the common standard) and define a quorum. This clause matters more than founders expect, because it’s the one auditors test directly against your calendar: if the charter says quarterly and you met twice last year, that’s an exception you wrote for yourself. Committee Structures Larger organizations delegate through audit, risk, and compensation committees, each with its own mini-charter. Smaller companies don’t need committees, but if your charter mentions them, they must exist and produce minutes. Never copy a public-company template with a phantom audit committee. Conflict of Interest Provisions A disclosure and recusal process for conflicts, usually paired with an annual attestation. This clause supports the independence claim: independence isn’t a one-time status, it’s maintained through disclosed and managed conflicts. Evidence of Board Member Expertise and Qualifications COSO’s “applies relevant expertise” point of focus means the board should be able to ask probing questions about security and risk, not just finance. Charters increasingly include a skills expectation clause, and

Most Drata reviews are written by Drata’s competitors. Scroll the first page of Google and you’ll find review posts from rival compliance platforms, each one ending with a pitch for their own tool. This one is different, and the bias runs the other way, so let’s put it on the table: Axipro is a Drata Gold Partner, and our consultants configure the platform for clients every week. That means we profit when companies choose Drata. It also means we know exactly where it saves you months, where the invoice grows faster than you planned, and when you should pick something else. This review covers all three. What Is Drata?​ Drata is a compliance automation platform (the industry calls the category GRC, for governance, risk, and compliance) founded in 2020 in San Diego by Adam Markowitz, Daniel Marashlian, and Troy Markowitz. Its core job: connect to your cloud infrastructure, identity provider, HR system, and code repositories, then continuously test your security controls against frameworks like SOC 2 and ISO 27001, collecting timestamped evidence as it goes. When your auditor shows up, most of the evidence is already packaged. Funding, Valuation, and Market Position Drata has raised $328 million, most recently a $200 million Series C in late 2022 that valued the company at $2 billion. It passed $100 million in annual recurring revenue in early 2025, acquired the trust center platform SafeBase for $250 million the same year, and now serves more than 8,000 customers. In late 2025 it earned a FedRAMP 20x Low Pilot Authorization, which puts it in a small group of compliance platforms cleared through the U.S. government’s modernized FedRAMP review track. Together with Vanta, it’s one of the two platforms almost every compliance buyer shortlists. Who Drata Is Built For The sweet spot is cloud-native companies from seed stage to mid-market: SaaS businesses pursuing their first SOC 2 or ISO 27001, and scaling teams juggling three or four frameworks at once. If your infrastructure lives in AWS, Azure, or GCP and your team uses standard tools like Okta, GitHub, and a mainstream HRIS, Drata’s automation covers a large share of your evidence collection out of the box. The further you drift from that profile (heavy on-prem systems, exotic tooling, air-gapped environments), the more manual work remains. How Drata Works: From Connection to Audit The workflow runs in five stages. First, you connect your tech stack through more than 270 native integrations covering cloud providers, identity, version control, HRIS, MDM, and ticketing. Second, continuous control monitoring kicks in: automated tests run around the clock against your connected systems, checking things like MFA enforcement, encryption settings, and access reviews. Third, automated evidence collection captures timestamped proof each time a test passes, building the evidence library your auditor will draw from. Fourth, when a test fails, remediation workflows and alerts route the issue to an owner through Slack, Jira, or email, with guidance on how to fix it. Fifth, the Audit Hub gives your auditor a scoped login to review evidence directly in the platform instead of trading spreadsheets and screenshots over email. In our client engagements, that last piece cuts back and forth more than any other feature. Auditors ask fewer clarifying questions when they can trace evidence to its source themselves. Drata’s Core Features Reviewed Overview of the Drata platform and compliance management dashboard. Multi-Framework Control Mapping Drata maintains a single control set mapped across every framework you activate. Pass an encryption control once, and it satisfies the corresponding requirements in SOC 2, ISO 27001, and HIPAA simultaneously. For multi-framework programs, this is the feature that pays for the platform. Adding ISO 27001 to an existing SOC 2 program typically starts you at 60 to 80 percent complete rather than zero. The Drata Agent The Drata Agent is a lightweight application installed on employee laptops. It checks device posture: screen lock, disk encryption, password manager, antivirus, OS updates. It reads configuration states, not files, browsing history, or keystrokes. Employees sometimes push back on installing it anyway, which is why we advise clients to communicate what it does and doesn’t see before rollout, not after the first complaint. Companies with an existing MDM like Jamf or Intune can often pull device evidence from that integration instead. Risk Management, Vendor Risk, and the Trust Center The built-in risk register lets you score risks by likelihood and impact and tie them to controls and remediation tasks. Vendor risk management got a genuine upgrade with the August 2025 agentic AI release, which now collects vendor evidence, reviews SOC 2 reports, and drafts risk summaries with far less manual chasing. The Trust Center, built on the acquired SafeBase product, gives you a public page where prospects can review your certifications and policies under NDA. Clients in active enterprise sales cycles tell us it measurably shortens security review, though note it’s a paid add-on at most tiers, not a bundled feature. Policies, Training, and the Rest Drata ships editable policy templates for every major framework, embedded security awareness training with completion tracking, and an API for anything the native integrations miss. The policy templates are a real accelerator for first-time programs, with one caveat we see constantly: teams accept templates wholesale without adapting them, then get flagged in audit when their actual practice doesn’t match their written policy. A template you don’t follow is worse than no template. Supported Compliance Frameworks Drata supports more than 30 frameworks. The ones that matter for most buyers: SOC 2 (Type I and Type II) against the AICPA Trust Services Criteria, ISO 27001, HIPAA (where Drata operationalizes safeguards, since no formal HIPAA certification exists), GDPR under the EU data protection rules, and PCI DSS. Coverage extends to CMMC, NIS2, DORA, FedRAMP, and various NIST standards. You can also build custom frameworks by mapping your own control set, useful for internal standards or customer-specific requirements. What Users Really Say Drata holds a 4.8 out of 5 on G2 across more than 1,100 reviews, the highest score among the

Vanta is worth it for most cloud-native companies chasing their first SOC 2 or ISO 27001. It’s a harder call if you run on-prem infrastructure, have unusual evidence requirements, or a budget that can’t absorb a renewal surprise. That’s the short answer. The longer one comes down to three things: how much of the platform’s automation applies to your stack, what the contract costs by year two, and how much compliance expertise you have in-house. This review draws on Vanta’s 2026 product releases, third-party procurement data, review platforms, and our experience at Axipro as a Vanta partner implementing the platform for clients across SOC 2, ISO 27001, and ISO 42001 engagements. We work inside the tool every week. We also see exactly where it stops working, and a human has to pick up. What Is Vanta? A Quick Overview​ Vanta is a compliance automation platform that now calls itself an Agentic Trust Platform. It connects to your cloud infrastructure, identity provider, code repositories, HR system, and device fleet, then runs continuous automated tests against the controls your target framework requires. It collects evidence on its own, maps it to controls, and packages the whole thing for your auditor. Vanta at a Glance Founded in 2018, Vanta now serves more than 15,000 customers, from early-stage startups to names like Atlassian, Duolingo, and Icelandair. The platform supports 35+ frameworks, ships 400+ integrations (the deepest library in the category), and runs over 1,400 pre-built automated tests. In 2026, Forrester named Vanta a Leader in The Forrester Wave: Governance, Risk, and Compliance Platforms, Q2 2026, the first time it appeared in the evaluation. Who Vanta Is Built For (Startups, Mid-Market, Enterprise) Startups remain the core market: roughly 58% of Vanta’s G2 reviews come from small businesses, typically SaaS companies that need a SOC 2 report to close their first enterprise deals. Mid-market teams use it to run multiple frameworks off shared evidence. The enterprise push is newer. In March 2026, Vanta shipped an Organizations Center and adaptive business unit scoping, which lets larger companies segment compliance by product, region, or team inside a single workspace instead of duplicating controls across accounts. Frameworks Vanta Supports Coverage includes SOC 2 (Type I and Type II), ISO 27001, ISO 42001 for AI management systems, HIPAA, GDPR, HITRUST, FedRAMP, PCI DSS, and the NIST AI RMF, among 35+ total. The AI governance coverage matters more each quarter: ISO 42001 and NIST AI RMF requests now show up in security questionnaires that had never mentioned AI before 2025. Vanta Key Features Reviewed Overview of the Vanta platform and compliance management dashboard.   Continuous Controls Monitoring This is the engine. Vanta’s 1,400+ tests run continuously against AWS, GCP, Azure, Okta, GitHub, and whatever else you’ve connected: are S3 buckets encrypted, is MFA enforced, are background checks done on time, does anyone hold access they shouldn’t? Failing controls get flagged with remediation guidance and SLA tracking, so compliance stops being an annual scramble and turns into something you maintain as you go. Automated Evidence Collection Instead of screenshots and spreadsheet exports, evidence flows in from your integrations and lands on the right controls. Cross-mapping is the underrated part: evidence you collect for SOC 2 gets reused for ISO 27001, HIPAA, or ISO 42001, which is why adding a second framework on Vanta takes weeks rather than months. The Vanta AI Agent (2026 Update) The AI Agent launched in mid-2025 and has moved fast since. In November 2025, Vanta rebuilt it as AI Agent 2.0, the core of the new Agentic Trust Platform, alongside a Risk Graph and Customer Commitments tracking. In March 2026, dedicated agents for compliance, third-party risk, and customer trust workflows. In June 2026, the Vanta Agent for Risk unified internal and vendor risk into one continuously updated view. In practice, the agent scans your program for inconsistencies, drafts policy change summaries for annual reviews, suggests control mappings when you upload policies, validates evidence before audits, and flags questionnaire gaps before they slow a security review. Vanta pitches it as a 24/7 GRC engineer. That’s marketing, but not empty marketing: it takes real hours of tedious work off your plate. Every draft still needs a human review before adoption, and the agent does its best work when a question maps to evidence you already hold. Insider Note: The AI Agent is only as good as its signal. If a large slice of your stack sits outside Vanta’s 400+ integrations, its suggestions shift from precise to generic. Test it against your actual environment during a trial, not a polished demo tenant. Policy, Vendor Risk, and Training Modules Policy templates cover the standard library, with AI-assisted drafting and version tracking. Vendor Risk Management (VRM) is a paid add-on that collects vendor evidence and generates AI risk summaries, feeding the broader third-party risk management picture. Security awareness training is built in, which removes one more standalone tool from the stack. Trust Center and Questionnaire Automation The Trust Center gives you a public page where prospects self-serve your security posture, and questionnaire automation drafts answers to inbound security reviews. Vanta reports automating over 80% of questionnaire responses with up to a 95% acceptance rate and 81% faster review completion. Those are vendor numbers, so apply a discount, but the direction matches what users report. Watch the caps: lower tiers limit automated questionnaires per year, and enterprise sales teams burn through those limits quickly. Access Reviews Access review campaigns pull directly from your identity provider, so quarterly reviews become a guided approval flow instead of a spreadsheet exercise. It’s a strong module; just know it sits in the Plus tier and above, not the entry plan. Vanta Pros and Cons (Honest Breakdown) Pros: Where Vanta Excels The integration library is the deepest in the category, and it shows during onboarding: most tests light up within days for a standard cloud stack. Auditor familiarity is a real, compounding advantage, since most CPA firms know Vanta’s exports and ask fewer clarification questions. Cross-framework evidence reuse

Learn how organizations can use the EU AI Act to build trust, speed up innovation, strengthen procurement, and gain a competitive advantage through effective AI governance. The Biggest Mistake Organizations Make About the EU AI Act When executives hear “EU AI Act,” their first thought is usually: another regulation, another compliance project, another expense. And who could blame them? Between GDPR, DORA, and NIS2, businesses are under real pressure to show they handle technology responsibly. Here’s what most of them miss, though. Complying with the EU AI Act does more than keep you clear of fines. Done well, it becomes a selling point. Companies that treat AI governance as a strategic skill earn customer trust and close enterprise deals faster. That matters because customers, investors, and regulators are asking tougher questions about AI than ever: Can you explain your AI decisions? How do you manage bias? Who’s accountable when something goes wrong? What controls protect sensitive data? The EU AI Act gives you a framework for answering them. When you can show good governance, you satisfy regulators, and you also win over the customers and partners deciding whether to trust you in the first place. That trust is worth real money in the AI era. And good governance doesn’t mean more bureaucracy. It means consistency. With clear ownership, defined risk processes, and transparent documentation, AI projects become easier to run and easier to scale. Teams stop reinventing governance for every new initiative and follow a repeatable framework instead, which speeds up decisions and cuts uncertainty. Companies with mature AI governance are already seeing this play out. They build more customer confidence in their AI products and answer procurement and due diligence requests in days instead of weeks. They run less risk of expensive AI failures or reputational damage, look credible to investors and regulators, and roll out AI consistently across the organization. In a market where trust increasingly drives purchasing decisions, that shows up in revenue. A McKinsey survey on the state of AI found that organizations investing in responsible AI practices are better positioned to capture value as adoption scales. Three Steps to Get Started You don’t need to transform the whole organization at once. Here’s where you can start: First, map your AI environment. Build an inventory of AI systems and know where they’re being used. You can’t govern what you can’t see, and most organizations are surprised by how many AI tools are already in play across teams. Second, spot risk early. Work out which high-risk use cases exist and build governance into the development lifecycle. The EU AI Act classifies systems by risk level, so knowing where your use cases fall tells you exactly how much scrutiny each one needs before it ships. Third, fold governance into existing processes. Add AI governance to your security, privacy, and enterprise risk programs instead of running it as a separate effort. This is where recognized standards like ISO/IEC 42001 pull their weight, giving you a structured management system that slots into what you already have rather than bolting on yet another silo. That’s enough to set your organization up for the long run. The debate around the EU AI Act shouldn’t be about staying on the right side of regulation. It should be about building AI that people trust and that you can actually scale. Organizations that put governance in place now will innovate faster and earn a stronger market reputation while their competitors scramble to catch up. In a few years, the edge will go to the companies that govern AI best, not the ones that use it the most. Is your organization preparing for the EU AI Act? Start by assessing your AI governance maturity and aligning your AI strategy with recognized frameworks like ISO/IEC 42001 and the NIST AI RMF, and turn compliance into a business advantage.

ISO 27001 for Startups

ISO/IEC 27001 certificates nearly doubled in a single year, from 48,671 in 2023 to 96,709 in 2024, according to ISO’s own certification survey. A big share of that jump comes from startups, not enterprises. The reason is simple: buyers stopped taking “we take security seriously” at face value, and a certificate is the fastest way to prove it.  This guide covers when a startup should pursue ISO 27001, what it costs, how long it takes, and how a small team gets certified without a dedicated security department. What Is ISO 27001 and Why It Matters for Startups ISO/IEC 27001 is the international standard for information security management. It doesn’t hand you a checklist of firewalls to buy. Instead, it asks you to build and run an Information Security Management System (ISMS): a documented, repeatable way of finding your security risks and doing something about them. Certification means an accredited third party checked that your ISMS works and matches the standard. For a startup, that distinction matters. You’re not being graded on whether you own expensive tools. You’re being graded on whether you can show a system, which is exactly what an enterprise buyer’s procurement team wants to see before they sign. The Core Principles: Confidentiality, Integrity, and Availability Everything in ISO 27001 traces back to the CIA triad: confidentiality, integrity, and availability. Confidentiality means only the right people see the data. Integrity means the data is accurate and hasn’t been tampered with. Availability means the data is there when someone needs it. Every control you put in place, and every risk you assess, ties back to protecting one of those three properties. ISO puts it plainly: an ISMS that meets the standard preserves the confidentiality, integrity, and availability of information by running a risk management process. Keep the triad in mind, and the rest of the framework stops feeling abstract. How ISO 27001 Differs from Other Security Frameworks for Early-Stage Companies SOC 2 is the framework startups usually bump into first, especially when selling into the US. It results in an attestation report from a CPA firm, scoped to specific systems. ISO 27001 is a certification, recognized in over 150 countries, and it covers your whole organization through a formal ISMS with management reviews and company-wide risk assessment. The two overlap heavily. Roughly 70 to 80 percent of the controls line up, so if you do one, the second gets much cheaper. The real difference is structure. SOC 2 checks whether specific controls work. ISO 27001 checks whether you’ve built a management system that keeps those controls working over time. It also aligns closely with GDPR, which is why it travels well in Europe. Insider Note: Auditors can usually tell within an hour whether your ISMS is real or was assembled the week before the audit. A management review meeting with actual notes, decisions, and follow-ups from three months ago is worth more than a perfect-looking policy binder with no evidence anyone ever used it. When Should a Startup Pursue ISO 27001 Certification? The honest answer: when a deal, a market, or an investor is asking for it, or is about to. Certifying purely because it feels responsible is a good way to burn cash and calendar time you don’t have yet. Early-Stage vs. Growth-Stage: Timing the Certification At pre-seed and seed, ISO 27001 is usually early unless you’re selling into regulated industries or the EU from day one. Your product and processes are still shifting, and certifying a moving target means re-documenting everything a quarter later. At Series A and beyond, the math changes. Deals get bigger, buyers get more careful, and investor due diligence starts probing your security posture. Certifying while you’re 15 to 40 people is often the sweet spot: mature enough to have stable processes, small enough that scoping the ISMS is still manageable. When ISO 27001 Might Be Overkill for Your Startup If your customers are US SMBs who only ever ask for SOC 2, leading with ISO 27001 may be solving a problem you don’t have. If you’re pre-revenue and still hunting for product-market fit, your time is better spent shipping. And if no one in your sales pipeline has ever mentioned a certificate, that silence is data. Pro Tip: Pull your Last 20 Security Questionnaires Before you commit, pull your last 20 security questionnaires or RFPs and count how many explicitly asked for ISO 27001 versus SOC 2 versus nothing. That single tally answers the “which framework, and when” question faster than any consultant’s discovery call. Key Benefits of ISO 27001 for Startups Unlocking Enterprise Sales and Bigger Deals The clearest return is revenue you couldn’t touch before. Large buyers often won’t even start a security review without a recognized certificate on file. ISO 27001 gets you past the first gate of enterprise sales, and it shortens the review itself because a big chunk of the questionnaire is already answered by your certification. Building Investor and Board Confidence Certification signals operational maturity. When an investor sees a functioning ISMS, they see a founder who can build systems, not only ship features. That plays well in investor due diligence, where a security gap can stall a term sheet, and it gives your board something concrete to point to on risk. Establishing Customer Trust from Day One A certificate is third-party proof, and third-party proof beats self-assurance every time. For a young company with no brand equity yet, it’s a shortcut to being taken seriously by customers who’ve never heard of you. Creating a Scalable Security Foundation Because ISO 27001 makes you build a system rather than a one-off fix, it scales as you grow. New hires, new products, and new data types slot into an ISMS you already run. You’re not rebuilding security from scratch at every stage. Reducing Long-Term Compliance Costs Adding SOC 2, HIPAA, or ISO 42001 later is far cheaper once an ISMS exists, thanks to that 70 to 80 percent control overlap. The first framework is the expensive one.

Compliance Hubs

Discover key insights, educational articles, helpful guides and more.

OWASP published the 2026 edition of its Top 10 for LLM Applications on August 4, 2026, during Black Hat week, and eight of the ten entries changed position. One got renamed. The message behind the reshuffle is blunt: you won’t build a model that can’t be fooled, so build the application around it in a way that limits the damage when it is. That one idea explains almost every move in the new ranking, and it should change how your team thinks about shipping AI features. This guide walks through the 2026 list in plain English: what each risk means, a real-world example, and what your team can actually do about it, with or without a dedicated security function. What Is the OWASP GenAI LLM Top 10 2026? The OWASP Top 10 for LLM Applications is a community-built awareness document that ranks the ten most critical security risks in applications powered by large language models. The OWASP GenAI Security Project, a global open-source initiative under the OWASP Foundation, maintains it, and the 2026 edition is the third release since the list first appeared in 2023. OWASP, the Open Worldwide Application Security Project, has published risk lists for web applications since 2003, and those lists became the shared vocabulary security teams, auditors, and buyers use to talk about risk. The GenAI LLM Top 10 does the same job for AI. Whether you’re a two-person startup wiring an API into a chatbot or an enterprise running retrieval pipelines, it gives you a common map of what actually goes wrong. One scoping note matters before anything else. The 2026 edition covers the model as a component inside an application: something that accepts input, generates output, and maybe retrieves information. The moment the model becomes an actor, with tools it can call and consequences it sets in motion, the risk shifts to the companion OWASP Top 10 for Agentic Applications from December 2025. Most products now do both, so most teams need both lists. Why the 2026 Update Matters for AI Builders Two things separate this edition from everything OWASP has published on AI so far. First, the methodology changed. Every previous version rested purely on expert consensus, meaning hundreds of practitioners voting on which risks matter most. This time the vote carried 75% of the weight, and the remaining 25% came from analysis of 6,639 real-world AI security incidents pulled from public vulnerability databases and an AI-harm database. It’s the first edition grounded in evidence of what has actually gone wrong rather than expert prediction of what might. Second, the framing changed. The project leads open the 2026 release by telling teams to stop optimizing the model and start optimizing the containment. The industry has spent two years pouring effort into filters, guardrail models, and jailbreak resistance. The 2026 list says: assume those will eventually fail, and make sure that when they do, nothing important breaks. AI security becomes blast radius control rather than perfect prevention. And this isn’t just a security engineer’s document. Developers decide what tools and permissions a model gets. Product owners decide which workflows run without a human in the loop. Founders and ops leads are the ones answering the security questionnaires where these questions now show up. The 2026 edition also ships a mapping appendix that connects every risk to frameworks your customers and auditors already recognize: NIST’s AI Risk Management Framework, MITRE ATLAS, MITRE CWE, and the Agentic Top 10. Insider Note: Enterprise vendor assessments have started asking about the OWASP LLM Top 10 by name. In security questionnaires we complete for clients at Axipro, questions like “describe your controls against prompt injection and excessive agency” began appearing in early 2026, sometimes before the buyer’s own team could explain what they meant. Being able to answer with a mapped control set is becoming a deal-cycle advantage, not just a security exercise. How the 2026 List Differs From Previous Versions The top two entries held their positions. Everything below them moved. Key Shifts Since the 2025 Update Excessive Agency jumped from sixth to third, the biggest promotion on the list. In 2025, giving a model tools and autonomy was mostly a theoretical worry. By 2026, agentic deployments had produced real production incidents, and the community concluded that agency is what decides whether a successful prompt injection is an inconvenience or a breach. Unbounded Consumption rose four places, from tenth to sixth. Inference costs became a real budget line as reasoning models, long outputs, and agent loops multiplied the compute behind a single request. “Denial of Wallet,” where an attacker spends pennies to trigger spend you can’t afford, is now a mainstream finding. Improper Output Handling fell from fifth to tenth. The risk didn’t shrink. It fell because it’s well understood and directly fixable with encoding and validation practices web developers already have. The entries above it are neither. What’s New, Renamed, or Reprioritized System Prompt Leakage became Hidden Context Exposure, and the scope widened a lot. The 2025 entry worried about attackers extracting your system prompt. The 2026 entry covers everything assembled into the model’s context that users aren’t meant to see: system instructions, retrieved policy documents, tool schemas, workflow rules. The guidance is unusually honest for a security document: assume all of it is discoverable, and design so that disclosure costs you nothing. Data and Model Poisoning absorbed fine-tuning subversion. The attack surface for corrupting a model’s behavior runs from pretraining data through fine-tuning pipelines into the retrieval stores RAG systems depend on, and the entry now says so. Misinformation climbed on evidence, not opinion. Practitioners voted it low; the incident data ranked it high. As reported in Help Net Security’s coverage of the release, OWASP also describes a “defense effect” working in the opposite direction on prompt injection: teams block it so effectively that few successful attacks reach public databases, which makes the risk look smaller than the money spent containing it. Signals About Where AI Security Is Heading Read together, the moves point one

For the past two years, enterprise AI risk conversations have centered on a familiar set of concerns: model bias, hallucination, data privacy, and dependency on third-party models. These are real risks, and most organizations now run some version of a governance program to manage them. But something has shifted. Organizations are no longer just deploying AI that generates content for a human to review. They’re deploying AI that acts. Agents now plan multi-step tasks, call APIs, move data between systems, execute transactions, and coordinate with other agents, often with no human checkpoint in the loop. That shift deserves more than a footnote in the existing AI risk category. It deserves its own line in the risk register: Agentic Autonomy Risk. What Is Agentic AI Risk Management? Agentic AI risk management is the practice of identifying, assessing, and controlling the risks created when AI systems take autonomous action on an organization’s behalf. Where traditional AI governance evaluates outputs (accuracy, bias, privacy), agentic AI risk management governs what agents actually do: the tools they call, the permissions they inherit, and the downstream consequences of their actions. That distinction is the reason existing risk registers struggle with agents, and it’s worth unpacking properly. What Agentic AI Actually Changes Traditional AI systems, even generative ones, are advisory. They produce an output such as a summary, a prediction, a draft email, or a classification, and a human remains the last checkpoint before anything happens in the real world. Agentic AI removes that checkpoint. An agentic system doesn’t just produce an answer. It pursues a goal. It decides which tools to call and in what order, then executes those actions directly against live systems: submitting a purchase order, modifying a database record, sending an external communication, or orchestrating a set of sub-agents to complete a broader workflow. Agentic autonomy is the degree to which a system can plan and execute actions without a human explicitly authorizing each step. It’s a spectrum rather than a binary. At one end, the AI drafts and a human approves every action. At the other, the AI operates within broad guardrails and only escalates exceptions. The further an organization moves along that spectrum, the less its exposure looks like software risk and the more it looks like delegated authority risk, the kind normally reserved for employees, contractors, and automated financial systems. Why Existing Risk Registers Miss Agentic AI Risks Most enterprise risk registers were built on a reasonably safe assumption: a human initiates consequential actions, and the technology around that human behaves deterministically. Agentic AI breaks both halves of that assumption at once. A few specific gaps show up quickly when organizations try to map agentic deployments onto existing categories. Operational risk registers assume process failures come from human error or system outages, not from a system independently choosing an unanticipated path to a stated goal. Cybersecurity risk registers are built around unauthorized external access, while an agent problem usually involves an authorized system taking unauthorized internal actions with its own legitimate credentials. Model risk frameworks, borrowed largely from financial services, evaluate output accuracy rather than action consequences, which matters most when those actions can’t be reversed. And third-party risk assessments treat vendors as static entities, not as autonomous agents that might invoke other vendors’ agents on your behalf. See our guide to the NIST AI Risk Management Framework for how output-focused frameworks are structured. The result is a governance blind spot. An organization can be compliant against its AI policy, its cybersecurity policy, and its vendor risk policy, and still have nobody accountable for the specific risk of a system initiating a harmful sequence of actions before anyone notices. Defining Agentic Autonomy Risk Agentic Autonomy Risk is the risk that an AI system, operating with delegated decision-making and execution authority, takes actions that are harmful, non-compliant, or misaligned with organizational intent before adequate human oversight can intervene. Those actions might happen independently or in coordination with other agents. It deserves standing as a named category alongside cybersecurity, operational, legal, financial, and third-party risk because the loss event itself is different. The harm is a completed action in a live system, and it may be difficult or impossible to reverse. The accountability structure is different too: when an orchestrating agent delegates to sub-agents, responsibility for the outcome gets distributed in ways existing ownership models don’t cleanly capture. So is the detection window. Traditional controls assume a human is positioned to catch an error before it compounds, but an agent can execute dozens of dependent actions faster than any human review cycle. 7 Agentic AI Risk Scenarios to Put on Your Register 1. Unauthorized autonomous decision-making. An agent takes an action within its technical permissions but outside its intended business mandate. It adjusts pricing, approves a refund, or modifies a customer record, and no policy ever explicitly authorized that scenario. 2. Goal misalignment. The agent optimizes for a literal interpretation of its objective in a way that diverges from actual business intent, particularly under ambiguous or adversarial inputs. 3. Multi-agent interactions and cascading failures. One agent’s flawed output becomes another agent’s trusted input. A single error can propagate across a chain of agents faster than anyone can detect it, amplifying the original mistake instead of containing it. 4. Excessive tool or system permissions. Agents get provisioned with broad, standing access “to be safe” rather than scoped, least-privilege access tied to specific tasks. A productivity tool quietly becomes a privilege-escalation path. 5. Regulatory non-compliance. Autonomous actions trigger obligations under data protection, financial services, employment, or sector-specific regulation, and they execute without the compliance review a human-initiated process would normally receive. 6. Explainability and accountability gaps. An autonomous action causes harm and the organization can’t clearly reconstruct why the agent chose that path, or establish whether the business owner, the AI governance function, or the vendor is accountable for the outcome. 7. Autonomous third-party actions. A vendor’s agent, integrated into your environment, takes action on your behalf, or your agent acts against a

A SOC 2 penetration test costs between $1,000 and $30,000 for most companies. A typical SaaS scope, meaning one web application, its API layer, and the cloud infrastructure behind it, usually lands between $2,000 and $20,000. Early-stage startups with a narrow scope can get an auditor-accepted test for $1,000 to $8,000, while enterprises with multiple products and hybrid infrastructure regularly spend $20,000 to $50,000 or more. The spread is wide because “penetration test” covers everything from an automated scan with a cover page to weeks of manual testing by senior engineers. Auditors know the difference, and so do the enterprise customers who asked for your SOC 2 report in the first place. This guide breaks down what drives the price, where the hidden costs sit, and how to buy a test that holds up in fieldwork without overpaying for it. What Is SOC 2 Penetration Testing?​ A SOC 2 penetration test is a simulated attack on your systems, performed by a qualified security professional, scoped to the environment covered by your SOC 2 report. The tester tries to exploit real weaknesses the way an attacker would: broken access controls, injection flaws, misconfigured cloud services, exposed credentials. The output is a report your auditor reads as evidence that your security controls work in practice, not only on paper. That last part matters. A pentest bought for SOC 2 has a second audience beyond your security team. If the report doesn’t map findings to your audit scope, document its methodology, and show remediation, it fails the job you bought it for. We cover the full deliverable in our guide to what a SOC 2-ready VAPT report includes. How Penetration Testing Fits Into SOC 2 Compliance​ SOC 2 is built on the AICPA’s Trust Services Criteria, and the Security category (the Common Criteria) applies to every report. Penetration testing is the standard way to satisfy CC7.1, which expects you to detect and monitor for new vulnerabilities, and it supports CC4.1, which covers ongoing evaluations of whether controls actually function. The AICPA’s points of focus explicitly mention vulnerability scanning and penetration testing as examples of how companies meet these criteria. In practice, the test slots into your audit timeline as an evidence item. Your auditor will ask for the report, check the test date against the audit period, and review how you handled the findings. Remediation is often scrutinized harder than the test itself, because it shows whether your vulnerability management process runs or merely exists. Is Penetration Testing Required for SOC 2?​ Strictly speaking, no. The Trust Services Criteria never use the word “mandatory” about penetration testing. You could theoretically satisfy CC7.1 with vulnerability scanning and strong monitoring alone. In reality, almost every auditor expects one, and skipping it invites two problems. First, your auditor may push back during fieldwork or add exceptions to the report. Second, the enterprise buyers reviewing your SOC 2 report increasingly look for pentest evidence specifically, and a report without it raises questions during procurement. Treat the test as effectively required and budget for it from the start of your SOC 2 compliance checklist. How Much Does SOC 2 Penetration Testing Cost? Typical Price Range for SOC 2 Pen Testing Most companies pay $1,000 to $30,000, with the median engagement for a SaaS business sitting around $12,000 to $15,000. Compliance-focused tests at the lower end of the market start around $1,000 to $5,000. Deep manual testing from established firms runs $10,000 to $30,000. Anything quoted below roughly $3,000 is almost certainly automated scanning packaged as a pentest, which auditors are getting better at spotting. Cost by Company Size (Startup, SMB, Enterprise) Company size is a proxy, not the driver. A 15-person company with three products and a legacy on-prem component will pay more than a 200-person company with one tightly scoped SaaS platform. Testers price effort, and effort follows scope. Cost by Test Type (Network, Web App, API, Cloud, Internal/External) Most SOC 2 engagements bundle two or three of these. The common package for a cloud-native SaaS company is web app plus API plus cloud configuration, which is why the $1,000 to $20,000 band comes up so often. Companies with office networks and internal systems in their audit scope add internal network testing, and the price climbs accordingly. Factors That Influence SOC 2 Penetration Testing Cost Scope and Number of Assets Tested Scope is the single biggest cost driver. Every additional application, API endpoint group, cloud account, or network segment adds testing hours. A pentest priced without a scoping call is a pentest priced on guesswork, and the guess usually favors the vendor. Complexity of Application or Infrastructure​ A simple CRUD app with two user roles tests quickly. A multi-tenant platform with role hierarchies, workflow engines, file processing, and third-party integrations takes far longer, because each of those features creates attack surface a tester has to work through manually. Authentication tiers matter especially: every distinct role needs testing for privilege escalation and cross-tenant data access. Testing Methodology (Black Box, Grey Box, White Box) Black box testing gives the tester nothing but a URL, grey box adds credentials and documentation, and white box adds source code and architecture diagrams. Grey box is the default for SOC 2 and usually the best value, since the tester spends time exploiting rather than discovering. White box costs more upfront but finds deeper issues. Black box sounds rigorous but often wastes paid hours on reconnaissance an attacker would run for free. Depth of Testing and Manual vs. Automated Approaches Automated scanning finds known vulnerability patterns. Manual testing finds business logic flaws, chained exploits, and authorization gaps that no scanner catches, and it’s the part auditors and security-literate customers actually value. The ratio of manual work to automation is the honest explanation for most price differences between two quotes covering the same scope. Tester Credentials and Firm Reputation Senior testers holding OSCP, GPEN, or CREST credentials bill higher rates, and firms with recognized methodologies charge a premium for the credibility their letterhead carries

Two compromised versions of LiteLLM sat on PyPI for roughly 40 minutes on the morning of March 24, 2026. That window was enough to capture secrets from around 434,000 CI/CD pipeline runs across nearly 2,500 organizations, including AWS, Samsung, Cisco, Salesforce, Siemens, and Deloitte. In August, researchers at CloudSEK and Hudson Rock confirmed they had obtained the raw exfiltrated data: a 153GB archive containing 433,909 files of environment variables, cloud keys, Kubernetes secrets, and API tokens harvested live from running pipelines, as covered by Help Net Security’s reporting on the credential archive. If LiteLLM runs anywhere in your stack, or you touch any AI proxy infrastructure at all, you need answers to three things: whether you were exposed, what to rotate first, and whether the rotation you did back in March actually held. That last one matters more than it sounds, because “we rotated everything” has already burned at least one very large company. How the Breach Happened The attack didn’t start with LiteLLM. On March 19, 2026, a threat group called TeamPCP compromised the build pipeline of Trivy, a vulnerability scanner half the industry runs, and pushed a poisoned release. LiteLLM’s own CI pipeline ran Trivy, so the poisoned scanner had legitimate read access to the project’s runner environment. The attackers used that to steal LiteLLM’s PyPI publishing tokens and ship two malicious releases of their own: versions 1.82.7 and 1.82.8. KICS and the Telnyx Python SDK got hit in the same campaign. The payload design is the part worth studying. The malicious package dropped a .pth startup hook into site-packages, so the code ran the moment any Python interpreter started on the machine, whether or not anything imported LiteLLM. From there it harvested environment variables, read local credential files like .aws/credentials and .kube/config, tried to move laterally across Kubernetes clusters, and installed a systemd backdoor dressed up as a generic telemetry service. InfoQ’s coverage of the PyPI compromise put downloads of the compromised release above 40,000. For scale, LiteLLM normally gets downloaded around 3 million times a day. The exfiltration had a nasty fallback, too. According to CloudSEK, stolen data was encrypted and sent to a typosquatted domain, and when that failed, the malware created a public repository inside the victim’s own GitHub account and uploaded the loot as a release asset. Some companies were publishing their own secrets to the open internet and had no idea. Worth Knowing: The malicious code only existed in the PyPI artifacts. The GitHub source repository stayed clean the whole time, so a developer reviewing the code on GitHub saw nothing wrong. Source review isn’t artifact verification. If you don’t check that what the registry serves matches the upstream source, this class of attack is invisible to you. How to Check If You Were Exposed Three checks, from quickest to most involved. 1. Confirm whether the compromised versions ever ran The malicious versions went live on PyPI at 10:39 UTC on March 24, 2026 and got quarantined about 40 minutes later. The project’s advice: treat any install from that day before 16:00 UTC as suspect. Search your lockfiles, pip caches, SBOMs, and container image histories for 1.82.7 and 1.82.8. And check your internal artifact mirrors. An Artifactory or Nexus proxy that cached the bad release in March can keep serving it internally long after PyPI pulled it. Keep the .pth mechanism in mind when you scope this. The question isn’t “which applications import LiteLLM,” it’s “which machines had the package installed at all,” because every Python process on an infected machine triggered the payload. 2. Hunt for persistence Rotation is pointless if the attacker still has a foothold. Check developer machines, CI runners, and containers for unauthorized .pth files in site-packages and for suspicious systemd units, especially anything posing as a system telemetry service. And review activity from March 24 onward, not just the 40-minute window. Persistence is there so the access outlives the infection. Pro Tip: Don’t limit the persistence hunt to live machines. Base container images rebuilt in late March may have baked the payload into every image derived from them since. Scan your image registry for the affected LiteLLM versions and for unexpected .pth files, then trace which running workloads came from flagged images. 3. Check whether your secrets are in the dump Hudson Rock has published a domain lookup tool and is running ethical disclosures for affected organizations, and CloudSEK maintains a high-confidence victim list. Use them, but know their limits. Attribution in this dataset is genuinely hard. One dump with a siriusxm.com committer email actually traced, through its self-hosted GitLab endpoints, to AdsWizz, a SiriusXM subsidiary. And a large share of the dumps are generic pipeline configurations with no identifying domain, email, or server name at all. Absence from a victim list is not evidence of absence. If your pipelines ran the compromised versions, assume exposure no matter what a lookup tool tells you. What to Rotate, in What Order The guidance from both research teams is blunt: treat every secret the LiteLLM environment could reach as compromised. That covers secrets on disk, in memory, injected into CI jobs, and anything retrievable through instance metadata services. Work down by blast radius: Priority Credential type Why it comes first 1 Cloud IAM keys (AWS, GCP, Azure) Direct control of infrastructure, data stores, and billing. This is where attackers monetize fastest. 2 GitHub and GitLab PATs, package publishing tokens These let an attacker poison your releases and turn your company into the next link in the supply chain. 3 Kubernetes service account tokens and kubeconfigs Lateral movement across clusters was built into the payload, not a theoretical risk. 4 Database passwords and third-party API keys Dumped in plain text in the archive, often with no attribution, so nobody will warn you they leaked. 5 AI provider API keys Billing abuse, quota theft, and access to whatever data flows through your LLM routing layer. One word matters more than the rest of this article: revoke, don’t just rotate. That

The EU AI Act names recruitment AI as high-risk. Annex III explicitly lists AI systems used for recruitment, candidate selection, and employment decisions, which pulls CV screeners, video interview platforms, and assessment tools into the most demanding compliance regime the Act contains. The original compliance date for these systems was August 2, 2026. In June 2026, the EU’s Digital Omnibus moved the deadline to December 2, 2027, a 16-month extension that has led many HR and talent teams to shelve the topic entirely. That’s a mistake, for two reasons. First, one rule that directly affects recruitment technology is already in force: the ban on emotion recognition in the workplace has applied since February 2, 2025, and it catches features still shipping in some video interview products today. Second, the deferred obligations didn’t shrink. Conformity assessments, human oversight design, bias monitoring, and documentation all still arrive in full, and the practical work of auditing a recruitment stack, renegotiating vendor contracts, and training hiring teams routinely takes a year or more. Here’s what the EU AI Act actually requires of employers and vendors using recruitment tools, on the timeline that now applies. Why Recruitment Tools Are Classified as High-Risk Under the EU AI Act​ Definition of High-Risk AI Systems in Hiring​ The Act takes a list-based approach. Annex III, point 4, designates as high-risk any AI system intended for the recruitment or selection of natural persons, including placing targeted job advertisements, analyzing and filtering applications, and evaluating candidates. The same point covers AI used for decisions on promotion, termination, task allocation, and monitoring of workers, so the classification follows the tool through the entire employment lifecycle, not just the hiring funnel. The reasoning is straightforward: hiring decisions shape access to livelihoods, and algorithmic discrimination in hiring is well documented. The European Commission’s regulatory framework for AI treats employment as one of the areas where an AI error or bias causes serious harm to fundamental rights. That’s the test for the high-risk tier. Types of Recruitment Tools Affected In practice, the high-risk classification captures most of the modern recruitment stack: CV and resume screeners that rank or filter applicants, video interview platforms that score responses or delivery, psychometric and skills assessment tools that produce scores feeding a hiring decision, sourcing and matching algorithms that decide which candidates a recruiter sees, and programmatic job ad targeting systems that determine who sees a vacancy at all. If the system’s output materially influences who advances and who does not, assume high-risk until proven otherwise. Important: Emotion recognition is not high-risk in the workplace. It is prohibited. Article 5 bans AI systems that infer emotions of people in the workplace (outside narrow medical and safety cases), and that ban has applied since February 2025 with the Act’s top penalty tier attached. If your video interview vendor markets “engagement scoring” or “sentiment analysis” of candidates, that feature needs to be switched off for EU hiring now, not in 2027. Recruitment Tools That May Fall Outside High-Risk Classification Not everything in the HR stack qualifies. The Act carves out systems performing narrow procedural tasks that do not materially influence decision outcomes. An applicant tracking system that stores applications, schedules interviews, and sends templated emails is a database with a workflow, not a high-risk AI system. The same goes for tools that transcribe interviews without scoring them, deduplicate candidate records, or generate first drafts of job descriptions for a human to edit. The line is decision influence: the moment a tool ranks, scores, filters, or recommends candidates, it crosses into Annex III territory. Deployers who rely on an exemption must be able to document that assessment, so “we decided it doesn’t count” needs to exist on paper. Extraterritorial Scope: Which Employers Are Covered The Act applies to providers placing AI systems on the EU market and to deployers established in the EU, but it also reaches further: it covers providers and deployers located outside the EU where the output of the system is used in the EU. For recruitment, the consequence is blunt. A US or UK company with no EU entity that uses an AI screener to filter applicants for roles based in Berlin or Dublin, or that screens candidates located in the EU, is using the system’s output in the Union. Brexit doesn’t move UK employers out of scope when they hire into or from the EU. Providers vs. Deployers of Recruitment AI Tools The Act splits obligations between the provider (the vendor that develops the tool and places it on the market) and the deployer (the employer using it). Most employers are deployers, and deployer obligations are lighter but real. One common trap: an employer that substantially modifies a high-risk system, or puts its own name on it, can be reclassified as a provider and inherit the full provider stack. Heavy customization of a screening model, or fine-tuning it on your own hiring data, can be enough to trigger this. Key Obligations for Employers Using AI Recruitment Tools Human Oversight in Automated Hiring Decisions Deployers must assign oversight of the system to people with the competence, training, and authority to intervene. That last word matters. A recruiter who rubber-stamps whatever the ranking algorithm produces, because nobody has time to review 800 rejected CVs, doesn’t count as oversight. Regulators and courts will look at whether the human could genuinely override the system and whether they ever did. Designing review checkpoints where a person can meaningfully change the outcome, and logging when they do, is the core of compliant deployment. Transparency Requirements Toward Candidates Employers must inform workers and their representatives before putting a high-risk AI system into use at work, and candidates subjected to such a system must be told it is being used. In countries with works councils, such as Germany, this obligation lands on top of existing co-determination rights, so employee representatives may need to be consulted before the tool goes live rather than just told afterward. Burying an AI disclosure in a privacy policy paragraph is unlikely to survive scrutiny.

A green dashboard is not an audit opinion. Compliance automation platforms like Vanta, Drata, Secureframe, and Hyperproof have made SOC 2 readiness faster and cheaper, but every audit cycle produces the same pattern: controls that sat at “passing” for months come back from the auditor with exceptions or requests for re-testing. The four controls below account for a disproportionate share of those rejections, and they all fail for the same underlying reason. The tool confirmed that evidence exists. The auditor tested whether the control actually operated. This article walks through each of the four: what auditors reject, why, and how to fix the evidence before fieldwork starts. Why Compliance Tools Show “Passing” But Auditors Still Reject Controls​ The Gap Between Automated Checks and Auditor Judgment Compliance platforms run continuous control monitoring: API calls that check whether a configuration exists, a document is uploaded, or a task is marked done. That’s real value. It catches drift, keeps evidence in one place, and saves weeks of screenshot collection. An audit is a different exercise. A SOC 2 examination is an attestation performed by a CPA firm under AICPA standards, and the auditor’s job is to form an independent opinion on whether your controls met the Trust Services Criteria. That opinion rests on professional judgment, not on whether an API integration returned a 200 response. What “Passing” Actually Means in Your Compliance Dashboard​ When a control shows “passing,” the platform is telling you one narrow thing: at the moment of the last scan, an automated test found the artifact or setting it was programmed to look for: MFA enforced in the identity provider, a policy document uploaded, a training campaign sitting at 100%. The test says nothing about whether the underlying process ran the way your control narrative claims it did, or whether it ran that way across the whole audit period. How Auditors Evaluate Controls Beyond the Checkbox Auditors test two dimensions. Design effectiveness asks whether the control, as described, would meet the criterion if it worked as intended. Operating effectiveness, the core of a SOC 2 Type 2 report, asks whether it actually did throughout the audit period. To answer that, the auditor pulls a population (every access review, every change, every new hire in the period), selects a sample, and inspects the evidence item by item. A dashboard status feeds into that process. It doesn’t replace it. Insider Note: Auditors increasingly ask for evidence outside the compliance platform precisely because they know what the platform auto-collects. If every artifact you produce comes from the same tool export, expect the auditor to independently pull the population from the source system and compare. Discrepancies between the two are one of the fastest routes to an exception. Control #1: Access Reviews That Automation Marks Complete but Auditors Reject Why Auditors Reject Automated Access Review Evidence​ User access reviews sit under the logical access criteria (CC6.1 through CC6.3), and they are the single most common source of audit exceptions we see. The typical failure: the platform generated a user list, someone clicked “complete,” and the dashboard turned green. The auditor then asks a simple question the evidence can’t answer: what did the reviewer actually decide? The Missing Element: Documented Reviewer Judgment​ An access review is a judgment control. Someone with knowledge of the system must look at each account and confirm the access is still appropriate for the person’s role. A timestamped task closure proves the task was closed. It doesn’t prove anyone assessed anything, and an “approve all” review completed in ninety seconds gets exactly the skepticism it deserves. What Auditors Actually Want to See in Access Review Evidence Auditors look for four things: The full population of accounts at the time of review (including service accounts and admin roles), Evidence of who reviewed it and when, explicit dispositions per account or group (retain, modify, revoke), and Proof that flagged access was actually removed. That last item, the deprovisioning ticket showing revocation within a defined window, is the piece most companies can’t produce. How to Fix Your Access Review Control Before the Audit​ Assign a named control owner per in-scope system, run reviews quarterly, and require reviewers to record a disposition for every line, not a blanket approval. When access is revoked, link the removal ticket to the review record. If a quarter was missed, don’t backfill it. Document it honestly and show the remediation, because auditors treat fabricated retroactive evidence far more severely than a disclosed gap. Control #2: Change Management Approvals That Pass Automated Scans​ Why Ticket Closure Isn’t Proof of Approval​ Change management (CC8.1) automation typically verifies that production changes link to a ticket and the ticket is closed. Auditors test something stricter: that each sampled change was approved by an authorized person before deployment. An approval added after the merge, or a ticket closed by the same engineer who wrote the code, fails that test even though every automated check came back green. The Segregation of Duties Problem Automation Misses Segregation of duties is the requirement that no single person can develop, approve, and deploy the same change. NIST’s SP 800-53 control catalog treats it as a foundational access control principle, and SOC 2 auditors apply the same logic. Small engineering teams trip on this constantly. Self-approved pull requests, admins who can bypass branch protection, direct pushes to main: a scanner sees “changes with tickets” while an auditor sees SoD violations. Emergency Changes and Retroactive Approvals: Common Rejection Triggers​ Every audit period contains hotfixes. Auditors don’t reject emergency changes. They reject emergency changes with no documented post-hoc review. If your policy says urgent changes get retroactive approval within two business days, the auditor will sample your emergency changes and check exactly that. No policy, or a policy nobody followed, produces an exception. Rebuilding Change Management Evidence Auditors Will Accept​ Enforce the control technically: branch protection requiring at least one independent reviewer, no admin bypass, and deploy pipelines that only run from protected branches. Then write the emergency change procedure down and generate the review artifact every time it fires.

Guides and Reports

Discover key insights, educational articles, helpful guides and more.
EU AI Act Hiring Map

AXIPRO STUDY New Study: Europe is hiring AI builders faster than AI governance professionals Axipro analyzed 3,519 AI-related job postings across eight EU countries. For every professional hired to keep AI lawful, safe and accountable, nearly seven were hired to build more of it, and the gap is widest exactly where you’d least expect. Take EU AI ACT READINESS QUIZZ 16 AI Builders : 1 AI Governors Sweden — Europe’s widest AI governance gap 3,519 Job Postings Analyzed 8 EU Countries 2 Role Categories: Builders vs Governors July 2026 Date of Job Postings Analyzed The findings Finding 1: Sweden hires 16 AI builders for every 1 person to govern them Throughout our data-set we found the same pattern across all eight countries: the more a nation hires to build AI, the less it hires to govern it. France runs eleven builders to every governor. Even Ireland, the most balanced in Europe, looks responsible mainly because the US tech giants headquartered there import global-governance discipline under overlapping DORA and AI Act pressure.  3.5→16 builders hired per governor, Europe’s most balanced country to its least. Ireland 3.5 Germany 5.7 Spain 6.0 Italy 7.1 Netherlands 7.2 Belgium 7.9 France 11.4 Sweden 16:1 0 4 8 12 16 Builders hired per AI governor Source: Axipro, 2026 Sweden has one of the strongest engineering cultures in Europe. It also carries the widest governance gap we measured: sixteen AI builders hired for every person hired to govern them. France sits close behind at eleven to one. The most balanced country, Ireland at 3.5 to one, looks responsible for a reason that has little to do with virtue. The US tech giants headquartered in Dublin import global governance discipline, and they do it under the combined weight of the AI Act and DORA, the EU financial-sector resilience regime in force since January 2025. Engineering strength does nothing to close a governance gap, and it may widen it. A country that ships AI faster produces more systems that fall under the Act’s scope and, on this evidence, fewer people positioned to document, monitor, and defend them. Being good at building AI offers no protection against governing it badly. The countries most confident in their technical talent are running the largest deficit against the law. Explore AI governance hiring by country Click any country to see how many AI builders it hires for every governance professional, and where it ranks against the rest of Europe. Germany — 5.7 builders per governorDE France — 11.4 builders per governorFR Spain — 6.0 builders per governorES Italy — 7.1 builders per governorIT Netherlands — 7.2 builders per governorNL Belgium — 7.9 builders per governorBE Ireland — 3.5 builders per governorIE Sweden — 16 builders per governorSE 3.5 — balanced 16 — widest gap Source: Axipro, 2026 Sweden 16builders for every governance professional Rank 1 of 8 · 20 governance roles vs 319 builder roles posted Only 30% of the AI governance roles name the AI Act Share this Embed this map Copy & paste — links back to Axipro Copy embed code Branded, one paste, backlink included. × Share this country insight Share this AI governance gap X / Twitter LinkedIn Facebook WhatsApp Bluesky Email Copy link Choose a platform or copy the link. A view of the same country-level dataset behind the interactive map: governance roles, builder roles, builder-to-governance ratio, and the share of governance postings that name the EU AI Act. AI governance jobs Europe statistics by country: governance roles, builder roles, builder-to-governance ratio and AI Act mention percentage. Country Governance roles Builder roles Builder-to-governance ratio AI Act mention % Sweden 20 319 16.0:1 30.0% France 39 443 11.4:1 38.5% Belgium 38 299 7.9:1 39.5% Netherlands 61 439 7.2:1 31.1% Italy 40 284 7.1:1 45.0% Spain 64 384 6.0:1 28.1% Germany 88 501 5.7:1 27.3% Ireland 96 335 3.5:1 14.6% Source: Axipro analysis of AI builder, governance and compliance job postings across eight European countries. “AI Act mention %” is the share of governance postings that explicitly name the EU AI Act. Finding 2: The law nobody names. Most AI governance jobs still do not mention the EU AI Act Europe spent years drafting the AI Act. It cleared the European Parliament, survived the Digital Omnibus revisions, and now carries penalties that reach €35 million or 7% of global turnover for the most serious breaches, a ceiling that makes GDPR fines look modest. Yet fewer than three in ten of the governance roles created to handle it actually name the law in the job description. Among builder roles, the figure collapses to one in twenty-five. More than 7 in 10 Governance job descriptions do not mention the EU AI Act. This number rises to 9 in 10 for all AI job descriptions. Despite hiring for governance, risk, privacy, and compliance roles, most employers are not yet translating the EU AI Act into explicit job requirements. That disconnect should stop you. The people being hired to make Europe compliant are, for the most part, not being hired against the Act by name. They are titled around adjacent ideas: risk, ethics, model validation, data protection. Some of that work will map onto the Act’s requirements. Much of it will not, because a role written without the regulation in view rarely produces the conformity assessments, technical documentation, and human-oversight structures the Act specifically demands. Readiness is even thinner than the headcount suggests. Simply counting governance hires overstates how many people are actually working the law. What job descriptions actually name The EU AI Act is visible in governance roles — but still absent from most job ads. Across the laws and frameworks most relevant to AI governance hiring, the EU AI Act appears in fewer than three in ten governance postings, and only 4% of builder postings. Law or framework Governance roles naming it Builder roles naming it All roles naming it Governance mentions EU AI Act 28.5% 4.0% 7.6% 127 GDPR 26.9% 5.7% 9.6% 120 ISO 27001 11.4% 1.3% 2.8% 51

SOC 2 Hub

All you need to know about SOC 2 compliance

A SOC 2 penetration test costs between $1,000 and $30,000 for most companies. A typical SaaS scope, meaning one web application, its API layer, and the cloud infrastructure behind it, usually lands between $2,000 and $20,000. Early-stage startups with a narrow scope can get an auditor-accepted test for $1,000 to $8,000, while enterprises with multiple products and hybrid infrastructure regularly spend $20,000 to $50,000 or more. The spread is wide because “penetration test” covers everything from an automated scan with a cover page to weeks of manual testing by senior engineers. Auditors know the difference, and so do the enterprise customers who asked for your SOC 2 report in the first place. This guide breaks down what drives the price, where the hidden costs sit, and how to buy a test that holds up in fieldwork without overpaying for it. What Is SOC 2 Penetration Testing?​ A SOC 2 penetration test is a simulated attack on your systems, performed by a qualified security professional, scoped to the environment covered by your SOC 2 report. The tester tries to exploit real weaknesses the way an attacker would: broken access controls, injection flaws, misconfigured cloud services, exposed credentials. The output is a report your auditor reads as evidence that your security controls work in practice, not only on paper. That last part matters. A pentest bought for SOC 2 has a second audience beyond your security team. If the report doesn’t map findings to your audit scope, document its methodology, and show remediation, it fails the job you bought it for. We cover the full deliverable in our guide to what a SOC 2-ready VAPT report includes. How Penetration Testing Fits Into SOC 2 Compliance​ SOC 2 is built on the AICPA’s Trust Services Criteria, and the Security category (the Common Criteria) applies to every report. Penetration testing is the standard way to satisfy CC7.1, which expects you to detect and monitor for new vulnerabilities, and it supports CC4.1, which covers ongoing evaluations of whether controls actually function. The AICPA’s points of focus explicitly mention vulnerability scanning and penetration testing as examples of how companies meet these criteria. In practice, the test slots into your audit timeline as an evidence item. Your auditor will ask for the report, check the test date against the audit period, and review how you handled the findings. Remediation is often scrutinized harder than the test itself, because it shows whether your vulnerability management process runs or merely exists. Is Penetration Testing Required for SOC 2?​ Strictly speaking, no. The Trust Services Criteria never use the word “mandatory” about penetration testing. You could theoretically satisfy CC7.1 with vulnerability scanning and strong monitoring alone. In reality, almost every auditor expects one, and skipping it invites two problems. First, your auditor may push back during fieldwork or add exceptions to the report. Second, the enterprise buyers reviewing your SOC 2 report increasingly look for pentest evidence specifically, and a report without it raises questions during procurement. Treat the test as effectively required and budget for it from the start of your SOC 2 compliance checklist. How Much Does SOC 2 Penetration Testing Cost? Typical Price Range for SOC 2 Pen Testing Most companies pay $1,000 to $30,000, with the median engagement for a SaaS business sitting around $12,000 to $15,000. Compliance-focused tests at the lower end of the market start around $1,000 to $5,000. Deep manual testing from established firms runs $10,000 to $30,000. Anything quoted below roughly $3,000 is almost certainly automated scanning packaged as a pentest, which auditors are getting better at spotting. Cost by Company Size (Startup, SMB, Enterprise) Company size is a proxy, not the driver. A 15-person company with three products and a legacy on-prem component will pay more than a 200-person company with one tightly scoped SaaS platform. Testers price effort, and effort follows scope. Cost by Test Type (Network, Web App, API, Cloud, Internal/External) Most SOC 2 engagements bundle two or three of these. The common package for a cloud-native SaaS company is web app plus API plus cloud configuration, which is why the $1,000 to $20,000 band comes up so often. Companies with office networks and internal systems in their audit scope add internal network testing, and the price climbs accordingly. Factors That Influence SOC 2 Penetration Testing Cost Scope and Number of Assets Tested Scope is the single biggest cost driver. Every additional application, API endpoint group, cloud account, or network segment adds testing hours. A pentest priced without a scoping call is a pentest priced on guesswork, and the guess usually favors the vendor. Complexity of Application or Infrastructure​ A simple CRUD app with two user roles tests quickly. A multi-tenant platform with role hierarchies, workflow engines, file processing, and third-party integrations takes far longer, because each of those features creates attack surface a tester has to work through manually. Authentication tiers matter especially: every distinct role needs testing for privilege escalation and cross-tenant data access. Testing Methodology (Black Box, Grey Box, White Box) Black box testing gives the tester nothing but a URL, grey box adds credentials and documentation, and white box adds source code and architecture diagrams. Grey box is the default for SOC 2 and usually the best value, since the tester spends time exploiting rather than discovering. White box costs more upfront but finds deeper issues. Black box sounds rigorous but often wastes paid hours on reconnaissance an attacker would run for free. Depth of Testing and Manual vs. Automated Approaches Automated scanning finds known vulnerability patterns. Manual testing finds business logic flaws, chained exploits, and authorization gaps that no scanner catches, and it’s the part auditors and security-literate customers actually value. The ratio of manual work to automation is the honest explanation for most price differences between two quotes covering the same scope. Tester Credentials and Firm Reputation Senior testers holding OSCP, GPEN, or CREST credentials bill higher rates, and firms with recognized methodologies charge a premium for the credibility their letterhead carries

A green dashboard is not an audit opinion. Compliance automation platforms like Vanta, Drata, Secureframe, and Hyperproof have made SOC 2 readiness faster and cheaper, but every audit cycle produces the same pattern: controls that sat at “passing” for months come back from the auditor with exceptions or requests for re-testing. The four controls below account for a disproportionate share of those rejections, and they all fail for the same underlying reason. The tool confirmed that evidence exists. The auditor tested whether the control actually operated. This article walks through each of the four: what auditors reject, why, and how to fix the evidence before fieldwork starts. Why Compliance Tools Show “Passing” But Auditors Still Reject Controls​ The Gap Between Automated Checks and Auditor Judgment Compliance platforms run continuous control monitoring: API calls that check whether a configuration exists, a document is uploaded, or a task is marked done. That’s real value. It catches drift, keeps evidence in one place, and saves weeks of screenshot collection. An audit is a different exercise. A SOC 2 examination is an attestation performed by a CPA firm under AICPA standards, and the auditor’s job is to form an independent opinion on whether your controls met the Trust Services Criteria. That opinion rests on professional judgment, not on whether an API integration returned a 200 response. What “Passing” Actually Means in Your Compliance Dashboard​ When a control shows “passing,” the platform is telling you one narrow thing: at the moment of the last scan, an automated test found the artifact or setting it was programmed to look for: MFA enforced in the identity provider, a policy document uploaded, a training campaign sitting at 100%. The test says nothing about whether the underlying process ran the way your control narrative claims it did, or whether it ran that way across the whole audit period. How Auditors Evaluate Controls Beyond the Checkbox Auditors test two dimensions. Design effectiveness asks whether the control, as described, would meet the criterion if it worked as intended. Operating effectiveness, the core of a SOC 2 Type 2 report, asks whether it actually did throughout the audit period. To answer that, the auditor pulls a population (every access review, every change, every new hire in the period), selects a sample, and inspects the evidence item by item. A dashboard status feeds into that process. It doesn’t replace it. Insider Note: Auditors increasingly ask for evidence outside the compliance platform precisely because they know what the platform auto-collects. If every artifact you produce comes from the same tool export, expect the auditor to independently pull the population from the source system and compare. Discrepancies between the two are one of the fastest routes to an exception. Control #1: Access Reviews That Automation Marks Complete but Auditors Reject Why Auditors Reject Automated Access Review Evidence​ User access reviews sit under the logical access criteria (CC6.1 through CC6.3), and they are the single most common source of audit exceptions we see. The typical failure: the platform generated a user list, someone clicked “complete,” and the dashboard turned green. The auditor then asks a simple question the evidence can’t answer: what did the reviewer actually decide? The Missing Element: Documented Reviewer Judgment​ An access review is a judgment control. Someone with knowledge of the system must look at each account and confirm the access is still appropriate for the person’s role. A timestamped task closure proves the task was closed. It doesn’t prove anyone assessed anything, and an “approve all” review completed in ninety seconds gets exactly the skepticism it deserves. What Auditors Actually Want to See in Access Review Evidence Auditors look for four things: The full population of accounts at the time of review (including service accounts and admin roles), Evidence of who reviewed it and when, explicit dispositions per account or group (retain, modify, revoke), and Proof that flagged access was actually removed. That last item, the deprovisioning ticket showing revocation within a defined window, is the piece most companies can’t produce. How to Fix Your Access Review Control Before the Audit​ Assign a named control owner per in-scope system, run reviews quarterly, and require reviewers to record a disposition for every line, not a blanket approval. When access is revoked, link the removal ticket to the review record. If a quarter was missed, don’t backfill it. Document it honestly and show the remediation, because auditors treat fabricated retroactive evidence far more severely than a disclosed gap. Control #2: Change Management Approvals That Pass Automated Scans​ Why Ticket Closure Isn’t Proof of Approval​ Change management (CC8.1) automation typically verifies that production changes link to a ticket and the ticket is closed. Auditors test something stricter: that each sampled change was approved by an authorized person before deployment. An approval added after the merge, or a ticket closed by the same engineer who wrote the code, fails that test even though every automated check came back green. The Segregation of Duties Problem Automation Misses Segregation of duties is the requirement that no single person can develop, approve, and deploy the same change. NIST’s SP 800-53 control catalog treats it as a foundational access control principle, and SOC 2 auditors apply the same logic. Small engineering teams trip on this constantly. Self-approved pull requests, admins who can bypass branch protection, direct pushes to main: a scanner sees “changes with tickets” while an auditor sees SoD violations. Emergency Changes and Retroactive Approvals: Common Rejection Triggers​ Every audit period contains hotfixes. Auditors don’t reject emergency changes. They reject emergency changes with no documented post-hoc review. If your policy says urgent changes get retroactive approval within two business days, the auditor will sample your emergency changes and check exactly that. No policy, or a policy nobody followed, produces an exception. Rebuilding Change Management Evidence Auditors Will Accept​ Enforce the control technically: branch protection requiring at least one independent reviewer, no admin bypass, and deploy pipelines that only run from protected branches. Then write the emergency change procedure down and generate the review artifact every time it fires.

Most SOC 2 preparation effort goes into access controls, encryption, and vendor reviews. Then the auditor’s first evidence request arrives, and item one has nothing to do with technology: show us your board charter, your meeting minutes, and proof that your board operates independently from management. That’s CC1.2, and it causes more last-minute scrambling than almost any technical control in the framework. This guide explains what CC1.2 requires, provides a board charter template with sample language that auditors accept, and covers the situation most startups actually face: satisfying the criterion without a traditional board of directors. What Is a SOC 2 Board Charter and Why It Matters for CC1.2 A board charter is a formal document that defines your board’s purpose, composition, authority, meeting procedures, and oversight responsibilities. Outside of compliance, it’s a corporate governance tool and a good idea in general for companies with shareholders.  Inside a SOC 2 audit, it’s the primary design evidence for CC1.2, the criterion that asks whether an independent body oversees management and the internal control environment. The charter matters because CC1.2 is one of the few criteria where the control is a document plus behavior. The charter establishes the structure. The auditor then tests whether the structure operates: did the board actually meet, did it review the security program, did it challenge management? A beautifully drafted charter with no meeting minutes behind it fails just as surely as no charter at all. If you’re earlier in your preparation, our complete SOC 2 guide covers how the full audit fits together. Understanding CC1.2: The Board Independence Criterion CC1.2 is part of the Trust Services Criteria published by the AICPA (American Institute of Certified Public Accountants). The criterion requires that the board of directors, in the AICPA’s words, “demonstrates independence from management and exercises oversight” of how internal control is developed and how it performs. That sentence hides two separate tests. Independence means the board isn’t just management wearing a second hat. Active oversight means the board actually reviews and challenges the control environment instead of existing on paper. Plenty of companies pass one and fail the other. How CC1.2 Fits Within the CC1 Control Environment The Common Criteria run from CC1 through CC9, and the CC1 series covers the control environment: the governance and people layer everything else rests on. CC1.1 addresses integrity and ethical values, CC1.2 addresses board independence and oversight, CC1.3 covers organizational structure and reporting lines, CC1.4 covers competence and hiring, and CC1.5 covers accountability. CC1.2 is the layer that makes the other four credible. A code of conduct means little if nobody independent of management ever checks whether leadership follows it. The COSO Principle 2 Connection The Trust Services Criteria are built directly on the COSO Internal Control—Integrated Framework and its 17 principles. CC1.2 maps to COSO Principle 2, which carries four points of focus: the board establishes oversight responsibilities, applies relevant expertise, operates independently of management, and provides oversight of the system of internal control. Those four phrases are worth memorizing, because they’re effectively the outline of a good board charter. Why Auditors Prioritize Board Charter Evidence Auditors test the control environment first because failures there cascade. If governance is weak, every other control claim gets harder to trust: who approved the risk assessment, who reviewed the incident report, who held management accountable when a control slipped? An exception at CC1.2 tells the auditor that nobody independent was watching, and they’ll read the rest of your evidence with that in mind. That’s why board charter requests sit near the top of almost every evidence list. Worth Knowing: Points of Focus Points of focus are not pass/fail requirements. The AICPA describes them as characteristics that assist evaluation, and the 2022 revisions changed points of focus without changing any criteria. In practice, though, they function as the auditor’s mental checklist, so drafting your charter against them is the safest move. What Auditors Actually Look For in a Board Charter Auditors don’t grade prose style. They scan for specific, verifiable commitments. Here’s what they check, roughly in order. Documented Board Independence from Management The charter must state how many members are independent, define what independence means (no operational role, no material financial relationship beyond board compensation or equity), and describe how independence is maintained. “The board includes members independent of management” without a definition is boilerplate; auditors want criteria they can test against actual member profiles. Defined Oversight Responsibilities This is the heart of CC1.2. The charter should explicitly assign the board oversight of internal control, information security, and risk management. If the charter only mentions financial oversight and strategy, it wasn’t written with SOC 2 in mind, and the auditor will notice the gap. Clear Authority and Decision-Making Powers What can the board approve, veto, or demand? Typical provisions include approving the risk management framework, reviewing audit results, approving executive appointments, and requiring management to report on control deficiencies. Authority without teeth reads as decorative. Meeting Cadence and Quorum Requirements The charter should commit to a minimum meeting frequency (quarterly is the common standard) and define a quorum. This clause matters more than founders expect, because it’s the one auditors test directly against your calendar: if the charter says quarterly and you met twice last year, that’s an exception you wrote for yourself. Committee Structures Larger organizations delegate through audit, risk, and compensation committees, each with its own mini-charter. Smaller companies don’t need committees, but if your charter mentions them, they must exist and produce minutes. Never copy a public-company template with a phantom audit committee. Conflict of Interest Provisions A disclosure and recusal process for conflicts, usually paired with an annual attestation. This clause supports the independence claim: independence isn’t a one-time status, it’s maintained through disclosed and managed conflicts. Evidence of Board Member Expertise and Qualifications COSO’s “applies relevant expertise” point of focus means the board should be able to ask probing questions about security and risk, not just finance. Charters increasingly include a skills expectation clause, and

SOC 2 and ISO 27001 Engagement

After a SOC 2 and ISO 27001 engagement, there are two documents out of the whole pile that actually close deals: the SOC 2 attestation report and the ISO 27001 certificate. Everything else your engagement produces exists to create those two, support them, or keep them alive for another year. Companies routinely ask their auditor for a SOC 2 certificate, which doesn’t exist. They send a prospect their full ISMS documentation when a one-page certificate would have done. They pay for six months of readiness work and then can’t say what they’re holding at the end of it. So here’s the full list. What a SOC 2 engagement produces, what an ISO 27001 engagement produces, what a combined program produces, and who gets to see each one. Understanding SOC 2 and ISO 27001 Engagement Outputs The Core Difference: Report vs. Certificate SOC 2 is an attestation. A licensed CPA firm examines your controls against the Trust Services Criteria under standards set by the AICPA, then writes up what it found and signs an opinion. No certificate. No logo from the AICPA. No pass or fail stamp. What you get is the report, and it usually runs 60 to 120 pages. ISO 27001 is a certification. An accredited certification body audits your Information Security Management System (ISMS) against ISO/IEC 27001:2022, and if you conform, it issues a certificate of registration. The certificate itself is a page or two. All the detail lives behind it, in your ISMS documentation and the audit reports the certification body writes as it goes. SOC 2 Engagement Deliverables The SOC 2 Attestation Report The report is the engagement. The AICPA’s illustrative SOC 2 report lays out the standard structure: auditor’s report, management’s assertion, system description, the Trust Services Criteria in scope, and the controls tested with their results. A Type I covers control design at one point in time. A Type II covers whether those controls actually operated over a period, usually three to twelve months, and most enterprise buyers now won’t accept anything else. Independent Auditor’s Opinion Letter First section of the report, and the first thing anyone experienced turns to. It gives the scope, the examination period, and the auditor’s conclusion. An unqualified opinion means the description held up and the controls worked. A qualified opinion means the auditor found something material, and every serious reviewer will want to talk about it. Management Assertion Your leadership signs a written statement stating that the system description is accurate and that the controls were properly designed and are operating. It reads like a formality, and it isn’t. The auditor’s entire examination runs against what management asserts here, so overstating anything creates real exposure. System Description Usually the longest part of the report, and you write it, not the auditor. It covers the services in scope, your infrastructure, software, people, processes, how data moves, which subservice organizations you depend on, and the complementary user entity controls your customers have to run on their side for your controls to hold up. Trust Services Criteria Applied Security (the Common Criteria) is in every SOC 2. Availability, Processing Integrity, Confidentiality, and Privacy are optional, and the report names exactly which ones you picked. Whatever you decide during scoping ends up printed in a document your customers read for the next several years. Description of Tests of Controls and Results (Type II) The matrix: every control, what the auditor did to test it, and what came back, including exceptions. Reviewers spend most of their time here, because the exceptions tell them things the opinion letter won’t. Bridge Letter / Gap Letter Your report covers a fixed window, so one ending December 31 leaves a hole for a customer doing diligence in June. A bridge letter from your management, not the auditor, confirms that nothing material changed in the control environment between the report’s end date and today. You’ll write these often enough to keep a template. Management Letter and Observations Plenty of auditors also send an internal-only letter covering observations, minor exceptions, and suggestions that never reached the threshold of a qualified opinion. It’s the closest thing to free consulting you’ll get before next year’s audit starts. Insider Note: Ask early whether your auditor issues a management letter, and whether exceptions land in the report body or only in that letter. Firms handle this differently, and the answer decides what your customers see versus what stays behind your firewall. It rarely comes up in the proposal, but it changes how the finished report reads to a buyer. ISO 27001 Engagement Deliverables ISO 27001 Certificate of Registration The document everyone asks for. It names the certified legal entity, states the ISMS scope, identifies the certification body, carries an accreditation mark from a body recognized under the International Accreditation Forum such as UKAS or ANAB, and shows the validity dates. It’s good for three years as long as you pass annual surveillance audits. Read the scope statement carefully, on your own certificate as much as anyone else’s. A certificate covering one office or one product line says nothing about the rest of the business. Statement of Applicability (SoA) After the certificate, this is the document buyers request most. The Statement of Applicability runs through all 93 Annex A controls in ISO/IEC 27001:2022, says which apply to you, justifies the ones you excluded, and records where each stands. Auditors use it as the map of your control environment, and larger customers increasingly want to see it or a summary of it during diligence. Risk Assessment and Risk Treatment Plan Your methodology, the register it produced, and the Risk Treatment Plan showing what you decided to do about each significant risk: mitigate it with a control, transfer it, avoid it, or accept it. ISO 27001 is built around risk, so these documents are what justify every control decision recorded in the SoA. Information Security Management System (ISMS) Documentation The policy and procedure set, plus the operational records that prove any of it happens. Information

SOC 2 Report vs. Certification

SOC 2 is not a certification, and no auditor will ever hand you a SOC 2 certificate. What you receive at the end of the audit is an attestation report: a detailed document, often 60 to 100 pages long, in which a licensed CPA firm expresses a professional opinion on your controls. That distinction sounds like pedantry until a prospect’s security team asks to see your “certificate” and you have nothing that looks like one. This article explains exactly what a SOC 2 report is, what it contains, how it differs from an ISO 27001 certificate, and how to talk about your SOC 2 status without misrepresenting it. Is SOC 2 a Certification or a Report? The Common Misconception About “SOC 2 Certification” Search volume tells the story: far more people look for “SOC 2 certification” than for “SOC 2 attestation,” and sales teams, procurement questionnaires, and even some auditors use the certification shorthand daily. The misconception is understandable. Every other major framework in the compliance stack, from ISO 27001 to PCI DSS, ends in something that looks like a pass. SOC 2 does not work that way, and treating it as if it does leads to awkward conversations during vendor due diligence. Why SOC 2 Is Technically an Attestation, Not a Certification A certification is a binary judgment issued by an accredited body: you meet the standard, or you do not. SOC 2 sits under the AICPA’s attestation standards, primarily SSAE 18 and its later amendments (SSAE 21 updated the relevant examination sections), specifically AT-C section 105 and AT-C section 205. Under those standards, an independent service auditor examines your controls and reports an opinion on them. Nobody “passes.” The auditor attests to what they found, in writing, with evidence. The output is a report, and the report is the entire deliverable. Understanding the SOC 2 Attestation Model What Is an Attestation Engagement? An attestation engagement is a formal examination in which a practitioner evaluates subject matter prepared by another party against defined criteria, then issues a written conclusion. In SOC 2, the subject matter is your system and its controls, the criteria are the AICPA’s Trust Services Criteria (Security, Availability, Processing Integrity, Confidentiality, and Privacy), and the party preparing the subject matter is you, the service organization. Security is the only mandatory category; the other four are scoped in based on your service commitments. The Role of the AICPA and Licensed CPA Firms The AICPA (American Institute of Certified Public Accountants) owns the SOC framework and the attestation standards behind it, but it does not perform audits and does not issue anything to your company. Only a licensed CPA firm can conduct a SOC 2 examination and sign the resulting opinion. That licensing requirement is the quality mechanism: the firm’s professional liability, independence rules, and peer review obligations stand behind the report. In practice, this means the assurance you get is only as strong as the auditor’s reputation and independence posture, which is why enterprise buyers often look at who signed the report almost as carefully as they look at what it says. How Attestation Differs from Certification and Accreditation The three terms describe different assurance models. Certification means an accredited certification body confirms conformity with a standard and issues a certificate, as happens with ISO 27001. Accreditation is one level up: it is the process by which national bodies, such as those coordinated through the International Accreditation Forum, authorize those certification bodies to certify in the first place. Attestation involves no certificate and no accreditation chain. A CPA firm examines evidence and expresses an opinion under professional standards. The credibility comes from the auditor’s license and independence, not from a badge. What You Actually Receive After a SOC 2 Audit The SOC 2 Attestation Report Explained The deliverable is a confidential, restricted-use document addressed to your management and intended for your customers, their auditors, and other informed parties. It is dense by design. A prospect’s risk team reads it to understand what your system does, which controls you operate, how the auditor tested them, and what the auditor found. It replaces a certificate with something far more useful: evidence. Key Components of the Final Report Independent service auditor’s opinion. The first section, usually two to three pages, states the auditor’s formal conclusion on whether your system description is fairly presented and whether your controls were suitably designed (and, for Type 2, operating effectively). This is the section report readers check first. Management’s assertion. A signed statement in which your leadership formally asserts that the system description is accurate and that controls meet the applicable criteria. SSAE 18 made this management assertion a mandatory element, which means responsibility for the description sits with you, not the auditor. System description. The longest narrative section was prepared by management against the AICPA’s SOC 2 description criteria. It covers the services provided, infrastructure, software, people, data, processes, subservice organizations, and complementary user entity controls. Trust Services Criteria and controls tested. A mapping of each in-scope criterion to the specific controls you operate. This is where scoping decisions become visible: a report covering Security only looks very different from one covering all five categories. Results of testing. For Type 2 reports, a control-by-control table showing the tests the auditor performed and the results, including any exceptions. Sophisticated readers spend most of their time here, because exceptions and the auditor’s response to them reveal more than the opinion page does. What a SOC 2 Report Is NOT (No Certificate, No Logo, No Pass/Fail Badge) There is no official SOC 2 certificate, no numbered credential, and no register of “certified” companies you can be listed in. The AICPA licenses a standard SOC logo that service organizations may display for a limited time after report issuance, but the logo confirms only that an examination took place. It says nothing about the opinion inside. Anyone selling you a “SOC 2 certificate” as a standalone artifact is selling something the framework does not produce. Important: If

SOC 2 Password Requirements

Most teams walk into a SOC 2 audit expecting standard requirements for their password policy: minimum length, 90-day rotation, one uppercase letter, one symbol, and so on. But there is no such checklist. The AICPA never published a list of mandatory password rules, and the federal guidance that most auditors lean on has thrown out half of what passed for best practice a decade ago.  Beyond compliance, this is remains a crucial cybersecurity control: Stolen and brute-forced credentials still drive a large share of breaches, and password policies are the main way to mitigate this risk. This guide covers what SOC 2 expects around passwords, where those expectations come from, and how to build a policy that satisfies an auditor without making your security worse. What Are SOC 2 Password Requirements? SOC 2 password requirements are the access controls that a service organization implements to govern how passwords are created, stored, enforced, and retired, all in service of the Trust Services Criteria. The important word is controls, not rules. SOC 2 does not hand you a specification. It asks whether your controls are suitably designed and operating effectively to keep unauthorized people out of your systems.   The Role of Passwords in the SOC 2 Trust Services Criteria The Trust Services Criteria, developed by the AICPA, are the evaluation standard for every SOC 2 report. Passwords sit inside the Security category, which is mandatory in all SOC 2 engagements, and specifically inside the Common Criteria series CC6, covering logical and physical access. Passwords are one of the most basic logical access controls you have, and one of the most scrutinized, because CC6 is usually the most evidence-intensive part of the entire audit. Relevant Common Criteria: CC6.1, CC6.2, and CC6.3 CC6.1 covers the controls that restrict logical access to systems, infrastructure, and data, this is where your password policy, MFA enforcement, and account lockout settings live. CC6.2 governs how access is granted, modified, and removed, meaning your provisioning workflows, access reviews, and offboarding processes are all evaluated here. CC6.3 focuses on the removal of access when it is no longer needed and the management of privileged credentials specifically. Together, these three criteria map to the full lifecycle of a credential: creation, ongoing use, and retirement. An auditor working through CC6 will expect evidence at every stage.   Does SOC 2 Mandate Specific Password Rules? No. The AICPA is explicit that the Trust Services Criteria do not define the controls an organization must have. You identify and implement controls that meet the criteria, and the auditor evaluates them. That means there is no AICPA-mandated minimum length, no required rotation interval, and no prescribed complexity formula. What the auditor checks is whether your stated controls exist, work, and reasonably prevent unauthorized access. Insider note: Auditors rarely fail you for choosing a 10-character minimum over 12. They fail you when your written policy says one thing and your actual system configuration says another. Consistency between the policy document and the enforced setting matters far more than the specific number. Why Password Requirements Matter for SOC 2 Compliance Preventing Unauthorized Access Credentials are the front door. The 2025 Verizon DBIR found that stolen credentials remained the single most common initial access vector, appearing in 22% of breaches, and that brute force attacks against basic web applications nearly tripled year over year. Strong authentication controls are the difference between an attacker hitting a wall and an attacker walking straight in with a valid login. Reducing Data Breach Risk Weak or reused passwords feed credential stuffing, where attackers replay username and password pairs harvested from earlier breaches against your login pages. Reuse is rampant: research from Microsoft’s Digital Defense Report routinely finds that the majority of people reuse passwords across services. A single leaked password elsewhere becomes a working key to your environment unless your controls catch it. Demonstrating Logical Access Controls to Auditors SOC 2 is an attestation. It is not enough to be secure; you have to prove it with evidence. Well-designed password controls produce exactly the artifacts an auditor wants: configuration screenshots, enforcement logs, MFA reports, and access review records. Good controls and good evidence are two sides of the same coin, and an internal audit process that routinely collects this evidence makes the formal engagement significantly less stressful. Core SOC 2 Password Requirements Although SOC 2 prescribes nothing specific, a defensible password policy almost always addresses the same set of controls. These are what auditors expect to see and what your peers in compliance treat as table stakes. Minimum Password Length Length is the strongest single lever for password entropy, and modern guidance favors it over everything else. A common defensible baseline is at least 12 characters for standard user accounts, with longer requirements for service and admin accounts. NIST SP 800-63B recommends that verifiers support passwords up to 64 characters so that passphrases and password-manager output are never truncated, an important implementation detail that many teams overlook. Password Complexity and Blocklists Old-style complexity rules, one uppercase, one symbol, one number, are fading, and for good reason. They push users toward predictable substitutions without meaningfully raising entropy. The more effective control is a blocklist: screening new passwords against dictionaries of common and previously breached credentials and rejecting matches. Tools like Have I Been Pwned’s Pwned Passwords API make this straightforward to implement. This stops Password1! from sneaking through even though it technically satisfies a legacy complexity rule. Password Rotation and History Forced periodic rotation is the control most teams keep out of habit, and it is also the one that modern guidance most clearly discourages. Rotation pushes users toward predictable patterns, Spring2025 becoming Summer2025, without improving security in any measurable way. Password history settings, which prevent the immediate reuse of recent passwords, still have a place, but blind calendar-based expiry should be replaced with event-driven resets: force a change when there is evidence of compromise, not because the calendar says 90 days have passed. Account Lockout After Failed Login Attempts An account

IAM Solutions Comparison 2026

The identity and access management market will pass $25 billion in 2026, and it is crowded with vendors that all make the same promise: the right people get the right access to the right resources at the right time. The hard part of any IAM solutions comparison is not finding capable products. It is that the leading platforms were each built to solve a different problem first, then expanded outward. Okta started with access. SailPoint started with governance. CyberArk started with privilege. Choose by brand reputation alone, and you risk buying a governance tool to solve an access problem, or paying enterprise prices for capabilities a mid-market team will never switch on. This guide compares the major providers by what they are actually good at, then walks through how to match one to your environment. What Is an IAM Solution? An IAM solution is the set of technologies that manages digital identities and controls what each identity can access. NIST frames the goal simply: ensure the right people and things have the right access to the right resources at the right time. In practice, that breaks into a few core functions: authenticating users (proving they are who they claim), authorizing them (deciding what they may do), and administering the account lifecycle as people join, move, and leave. The category splits into recognizable disciplines. Access management (AM) handles authentication and single sign-on. Identity governance and administration (IGA) handles who should have access and proves it to auditors. Privileged access management (PAM) protects the high-value accounts that can change infrastructure or read sensitive data. Most vendors now sell across these lines, but few are equally strong in all of them. That gap is the whole reason a comparison is worth doing. Why Comparing IAM Solutions Matters in 2026 Identity is now the primary attack surface. Stolen credentials and phishing remain among the top routes attackers use to get inside, which is why identity spending keeps climbing even when other security budgets flatten. The IAM market reached roughly $22 billion in 2025 and is on track for about $25 billion in 2026, growing near 15 percent a year, according to Fortune Business Insights. Two shifts make the comparison harder than it was a few years ago. First, the workforce went hybrid and cloud-first, so identity has to span on-prem systems, SaaS, and multi-cloud at once. Second, machine identities exploded. Your choice of platform now locks in how well you can govern not just employees but the service accounts, tokens, and AI agents multiplying across your environment. Gartner has reported that roughly 48 percent of organizations still lack a written IAM strategy — a serious problem, because a comparison is worth little if it is not anchored to documented requirements. Vendor demos are designed to make every product look like the obvious answer. Key Criteria for Comparing IAM Solutions A useful comparison rests on a consistent scorecard rather than the feature checklists vendors supply. The criteria below are the ones that tend to decide satisfaction two years after purchase. Core Identity and Access Capabilities Start with the fundamentals: single sign-on, multi-factor authentication, lifecycle provisioning and deprovisioning, and access certification. The differentiator in 2026 is adaptive, risk-based authentication that weighs device, location, and behavior before granting access, alongside phishing-resistant methods such as passkeys. A tool that only does password-plus-OTP is already behind. Deployment Options: Cloud-Native, Hybrid, and On-Premises Deployment model shapes cost, speed, and control. Cloud-native SaaS platforms deploy fastest and shift maintenance to the vendor. On-prem suits organizations with strict data-residency rules or deep legacy systems. Hybrid is the common reality, and the question to ask is how gracefully a platform bridges old and new — not whether it claims to. Integration Capabilities with Existing Infrastructure An IAM platform is only as good as its connectors. Look for prebuilt integrations with your core systems, directory services, HR platforms, and major SaaS apps, plus open standards support: SAML, OIDC, SCIM, and increasingly standards for continuous authorization. A thin connector catalog means custom engineering, which is where budgets quietly disappear. Scalability for Enterprise vs. Mid-Market Organizations Scale is not only user count. It is the number of applications, directories, and identity types a platform can govern without performance or administrative strain. Enterprise suites assume a dedicated identity team. Mid-market tools assume a stretched IT generalist. Buying the wrong tier means either paying for unused complexity or hitting a ceiling within two years. Pricing Models and Total Cost of Ownership Headline per-user pricing rarely reflects real cost. Implementation, professional services, connector licensing, premium support, and the internal staff time to run the platform often exceed the subscription itself. Compliance and Audit Support For regulated industries, audit support is a core feature, not a bonus. Strong platforms run access certification campaigns, segregation-of-duties checks, and audit-ready reports aligned with frameworks such as SOX, HIPAA, ISO 27001, and PCI DSS. The NIST Digital Identity Guidelines (SP 800-63, revised in 2025) are a useful reference for the assurance levels your authentication should meet. Vendor Support, Stability, and Roadmap You are buying a multi-year relationship. Financial stability, support quality, and a credible roadmap matter as much as today’s feature set, especially as the market consolidates and converges. A vendor that gets acquired or pivots can leave you maintaining a product on a slow decline. Pro Tip: Comparing Quotes When you compare quotes, normalize them to a three-year total cost of ownership that includes implementation and at least one major version upgrade. Vendors that look cheap per seat sometimes carry the heaviest services bill, and the gap usually shows up in year one, not at signing. IAM Solutions Compared: The Leading Providers The vendors below dominate enterprise shortlists. Each entry notes the problem the platform solves best — which is the most reliable way to read past the marketing. Okta Workforce Identity Cloud Okta is the largest independent identity vendor and was named a Leader in the 2025 Gartner Magic Quadrant for Access Management for the ninth straight year. Its strength is breadth of

Secure Data Disposal ISO 27001 SOC 2 GDPR

Researchers who buy second-hand drives off online marketplaces keep finding the same thing: live data.  A widely cited study by Blancco Technology Group found that 42% of used drives sold on eBay still held recoverable information, including financial records and personal data the previous owners assumed was long gone. The drives were not hacked; they were thrown away by organizations that treated deleting a file as the same thing as destroying it. Secure data disposal is where many compliance programs fail. ISO 27001, SOC 2, and GDPR all demand it, but they describe it in different languages, enforce it through different mechanisms, and punish failure in very different ways.  This article sets out what each framework requires, where the requirements overlap, and how to run a single disposal program that satisfies all three at once. Why Secure Data Disposal Matters Across Compliance Frameworks Disposal is the last link in the data lifecycle, and the easiest one to skip. An organization can run flawless access controls, encryption, and monitoring for years and still cause a reportable breach the moment one unwiped laptop leaves the building. A recoverable drive in a recycling skip is functionally identical to an open database on the internet, and auditors and regulators know it. Most disposal failures are unforced errors: a control that was already written into policy but never carried through to the actual hardware. The gap between having a disposal policy and proving this specific drive was destroyed is exactly where audits and breach investigations live. Defining Secure Data Disposal: Key Terms and Concepts What Is Secure Data Disposal? Secure data disposal is the end-to-end process of removing data and the equipment that holds it from active use, in a way that prevents its recovery. It covers the full lifecycle end: deletion of data while a system is still live, sanitisation of media that will be reused, physical destruction of media that will not, and the safe handling of equipment that is recycled, returned to a lessor, or sold. Disposal is the goal. The methods are how you get there. What Is Secure Data Destruction? Secure data destruction is the subset of disposal that renders media permanently unusable or its contents mathematically irretrievable. Shredding a drive, pulverising it, incinerating it, or destroying the encryption keys that make an encrypted disk readable are all forms of destruction. Destruction is one route to disposal, and it is the right route when the data is highly sensitive, or the media will never be reused. Secure Data Disposal vs. Secure Data Destruction: What Is the Difference? The distinction matters more than it looks. Disposal is the outcome you owe to every framework: data gone, unrecoverable, equipment handled appropriately. Destruction is just one of the methods. You can dispose of data without destroying the hardware by sanitising a drive thoroughly enough to reuse it. Confusing the two leads to two classic mistakes: destroying assets that could have been securely wiped and reused, and assuming a quick deletion counts as disposal when it does not. Important: Emptying the recycle bin, formatting a drive, or hitting delete does not dispose of data under any of these frameworks. Standard deletion only removes the pointer to the data; the bits remain until they are overwritten. Every framework discussed here expects the data to be unrecoverable, which is a far higher bar than not visible. What ISO 27001 Requires for Secure Data Disposal ISO/IEC 27001 handles disposal through a small cluster of Annex A controls that auditors read as a single process rather than in isolation. The two controls that do most of the work are 7.14 and 8.10. For a deeper look at how these controls fit into a broader compliance program, see our ISO 27001 implementation guide. ISO 27001 Annex A 7.14: Secure Disposal or Re-Use of Equipment Annex A 7.14 is a physical control. Before any equipment is disposed of or reused, the organisation must check whether it holds information assets or licensed software and ensure those are permanently erased or the media physically destroyed. It applies to servers, laptops, desktops, mobile devices, printers, network gear, and any storage media: if it ever processed information, it is in scope. The control replaces the older 2013 clause 11.2.7 and adds explicit expectations around removing identifying markings and handling end-of-occupancy scenarios. ISO 27001 Control 8.10: Information Deletion Annex A 8.10 is a technological control, and it focuses on the data rather than the box. It requires information stored in systems, devices, or media to be deleted when it is no longer required, and rendered unrecoverable. The cleanest way to keep these straight: 8.10 governs the data while it is in use or reaches its retention limit; 7.14 governs the hardware at end of life. Most retention-driven deletion sits under 8.10; most decommissioning sits under 7.14. ISO 27001 Control 8.12: Data Leakage Prevention and Its Role in Disposal Control 8.12 is rarely filed under disposal, but improperly discarded media is one of the oldest data leakage channels there is. A drive that leaves your control with recoverable data on it is a leak, regardless of how it left. Treating disposal as part of your leakage prevention posture forces the right question at the right time: what could walk out the door on this device, and has it actually been removed? Physical Destruction and Irretrievable Erasure Under ISO 27001 ISO 27001 offers two broad routes: physically destroy media that holds information, or erase and overwrite it so retrieval by a malicious party is precluded. The standard cross-references ISO/IEC 27040 for detailed sanitisation methods. The unifying requirement is that recovery should be impractical, not merely inconvenient. Deletion alone never satisfies this. Overwriting, Full-Disk Encryption, and Other Approved Methods Overwriting user-accessible storage with multiple passes is acceptable for many sensitivity levels. Full-disk encryption changes the economics of disposal entirely: if a device is encrypted from day one and the keys are properly managed, secure disposal can be as simple as destroying the keys, a technique known as

ISO 27001 Hub

The latest resources and guides for ISO 27001 Certification

ISO 27001 for Startups

ISO/IEC 27001 certificates nearly doubled in a single year, from 48,671 in 2023 to 96,709 in 2024, according to ISO’s own certification survey. A big share of that jump comes from startups, not enterprises. The reason is simple: buyers stopped taking “we take security seriously” at face value, and a certificate is the fastest way to prove it.  This guide covers when a startup should pursue ISO 27001, what it costs, how long it takes, and how a small team gets certified without a dedicated security department. What Is ISO 27001 and Why It Matters for Startups ISO/IEC 27001 is the international standard for information security management. It doesn’t hand you a checklist of firewalls to buy. Instead, it asks you to build and run an Information Security Management System (ISMS): a documented, repeatable way of finding your security risks and doing something about them. Certification means an accredited third party checked that your ISMS works and matches the standard. For a startup, that distinction matters. You’re not being graded on whether you own expensive tools. You’re being graded on whether you can show a system, which is exactly what an enterprise buyer’s procurement team wants to see before they sign. The Core Principles: Confidentiality, Integrity, and Availability Everything in ISO 27001 traces back to the CIA triad: confidentiality, integrity, and availability. Confidentiality means only the right people see the data. Integrity means the data is accurate and hasn’t been tampered with. Availability means the data is there when someone needs it. Every control you put in place, and every risk you assess, ties back to protecting one of those three properties. ISO puts it plainly: an ISMS that meets the standard preserves the confidentiality, integrity, and availability of information by running a risk management process. Keep the triad in mind, and the rest of the framework stops feeling abstract. How ISO 27001 Differs from Other Security Frameworks for Early-Stage Companies SOC 2 is the framework startups usually bump into first, especially when selling into the US. It results in an attestation report from a CPA firm, scoped to specific systems. ISO 27001 is a certification, recognized in over 150 countries, and it covers your whole organization through a formal ISMS with management reviews and company-wide risk assessment. The two overlap heavily. Roughly 70 to 80 percent of the controls line up, so if you do one, the second gets much cheaper. The real difference is structure. SOC 2 checks whether specific controls work. ISO 27001 checks whether you’ve built a management system that keeps those controls working over time. It also aligns closely with GDPR, which is why it travels well in Europe. Insider Note: Auditors can usually tell within an hour whether your ISMS is real or was assembled the week before the audit. A management review meeting with actual notes, decisions, and follow-ups from three months ago is worth more than a perfect-looking policy binder with no evidence anyone ever used it. When Should a Startup Pursue ISO 27001 Certification? The honest answer: when a deal, a market, or an investor is asking for it, or is about to. Certifying purely because it feels responsible is a good way to burn cash and calendar time you don’t have yet. Early-Stage vs. Growth-Stage: Timing the Certification At pre-seed and seed, ISO 27001 is usually early unless you’re selling into regulated industries or the EU from day one. Your product and processes are still shifting, and certifying a moving target means re-documenting everything a quarter later. At Series A and beyond, the math changes. Deals get bigger, buyers get more careful, and investor due diligence starts probing your security posture. Certifying while you’re 15 to 40 people is often the sweet spot: mature enough to have stable processes, small enough that scoping the ISMS is still manageable. When ISO 27001 Might Be Overkill for Your Startup If your customers are US SMBs who only ever ask for SOC 2, leading with ISO 27001 may be solving a problem you don’t have. If you’re pre-revenue and still hunting for product-market fit, your time is better spent shipping. And if no one in your sales pipeline has ever mentioned a certificate, that silence is data. Pro Tip: Pull your Last 20 Security Questionnaires Before you commit, pull your last 20 security questionnaires or RFPs and count how many explicitly asked for ISO 27001 versus SOC 2 versus nothing. That single tally answers the “which framework, and when” question faster than any consultant’s discovery call. Key Benefits of ISO 27001 for Startups Unlocking Enterprise Sales and Bigger Deals The clearest return is revenue you couldn’t touch before. Large buyers often won’t even start a security review without a recognized certificate on file. ISO 27001 gets you past the first gate of enterprise sales, and it shortens the review itself because a big chunk of the questionnaire is already answered by your certification. Building Investor and Board Confidence Certification signals operational maturity. When an investor sees a functioning ISMS, they see a founder who can build systems, not only ship features. That plays well in investor due diligence, where a security gap can stall a term sheet, and it gives your board something concrete to point to on risk. Establishing Customer Trust from Day One A certificate is third-party proof, and third-party proof beats self-assurance every time. For a young company with no brand equity yet, it’s a shortcut to being taken seriously by customers who’ve never heard of you. Creating a Scalable Security Foundation Because ISO 27001 makes you build a system rather than a one-off fix, it scales as you grow. New hires, new products, and new data types slot into an ISMS you already run. You’re not rebuilding security from scratch at every stage. Reducing Long-Term Compliance Costs Adding SOC 2, HIPAA, or ISO 42001 later is far cheaper once an ISMS exists, thanks to that 70 to 80 percent control overlap. The first framework is the expensive one.

SOC 2 and ISO 27001 Engagement

After a SOC 2 and ISO 27001 engagement, there are two documents out of the whole pile that actually close deals: the SOC 2 attestation report and the ISO 27001 certificate. Everything else your engagement produces exists to create those two, support them, or keep them alive for another year. Companies routinely ask their auditor for a SOC 2 certificate, which doesn’t exist. They send a prospect their full ISMS documentation when a one-page certificate would have done. They pay for six months of readiness work and then can’t say what they’re holding at the end of it. So here’s the full list. What a SOC 2 engagement produces, what an ISO 27001 engagement produces, what a combined program produces, and who gets to see each one. Understanding SOC 2 and ISO 27001 Engagement Outputs The Core Difference: Report vs. Certificate SOC 2 is an attestation. A licensed CPA firm examines your controls against the Trust Services Criteria under standards set by the AICPA, then writes up what it found and signs an opinion. No certificate. No logo from the AICPA. No pass or fail stamp. What you get is the report, and it usually runs 60 to 120 pages. ISO 27001 is a certification. An accredited certification body audits your Information Security Management System (ISMS) against ISO/IEC 27001:2022, and if you conform, it issues a certificate of registration. The certificate itself is a page or two. All the detail lives behind it, in your ISMS documentation and the audit reports the certification body writes as it goes. SOC 2 Engagement Deliverables The SOC 2 Attestation Report The report is the engagement. The AICPA’s illustrative SOC 2 report lays out the standard structure: auditor’s report, management’s assertion, system description, the Trust Services Criteria in scope, and the controls tested with their results. A Type I covers control design at one point in time. A Type II covers whether those controls actually operated over a period, usually three to twelve months, and most enterprise buyers now won’t accept anything else. Independent Auditor’s Opinion Letter First section of the report, and the first thing anyone experienced turns to. It gives the scope, the examination period, and the auditor’s conclusion. An unqualified opinion means the description held up and the controls worked. A qualified opinion means the auditor found something material, and every serious reviewer will want to talk about it. Management Assertion Your leadership signs a written statement stating that the system description is accurate and that the controls were properly designed and are operating. It reads like a formality, and it isn’t. The auditor’s entire examination runs against what management asserts here, so overstating anything creates real exposure. System Description Usually the longest part of the report, and you write it, not the auditor. It covers the services in scope, your infrastructure, software, people, processes, how data moves, which subservice organizations you depend on, and the complementary user entity controls your customers have to run on their side for your controls to hold up. Trust Services Criteria Applied Security (the Common Criteria) is in every SOC 2. Availability, Processing Integrity, Confidentiality, and Privacy are optional, and the report names exactly which ones you picked. Whatever you decide during scoping ends up printed in a document your customers read for the next several years. Description of Tests of Controls and Results (Type II) The matrix: every control, what the auditor did to test it, and what came back, including exceptions. Reviewers spend most of their time here, because the exceptions tell them things the opinion letter won’t. Bridge Letter / Gap Letter Your report covers a fixed window, so one ending December 31 leaves a hole for a customer doing diligence in June. A bridge letter from your management, not the auditor, confirms that nothing material changed in the control environment between the report’s end date and today. You’ll write these often enough to keep a template. Management Letter and Observations Plenty of auditors also send an internal-only letter covering observations, minor exceptions, and suggestions that never reached the threshold of a qualified opinion. It’s the closest thing to free consulting you’ll get before next year’s audit starts. Insider Note: Ask early whether your auditor issues a management letter, and whether exceptions land in the report body or only in that letter. Firms handle this differently, and the answer decides what your customers see versus what stays behind your firewall. It rarely comes up in the proposal, but it changes how the finished report reads to a buyer. ISO 27001 Engagement Deliverables ISO 27001 Certificate of Registration The document everyone asks for. It names the certified legal entity, states the ISMS scope, identifies the certification body, carries an accreditation mark from a body recognized under the International Accreditation Forum such as UKAS or ANAB, and shows the validity dates. It’s good for three years as long as you pass annual surveillance audits. Read the scope statement carefully, on your own certificate as much as anyone else’s. A certificate covering one office or one product line says nothing about the rest of the business. Statement of Applicability (SoA) After the certificate, this is the document buyers request most. The Statement of Applicability runs through all 93 Annex A controls in ISO/IEC 27001:2022, says which apply to you, justifies the ones you excluded, and records where each stands. Auditors use it as the map of your control environment, and larger customers increasingly want to see it or a summary of it during diligence. Risk Assessment and Risk Treatment Plan Your methodology, the register it produced, and the Risk Treatment Plan showing what you decided to do about each significant risk: mitigate it with a control, transfer it, avoid it, or accept it. ISO 27001 is built around risk, so these documents are what justify every control decision recorded in the SoA. Information Security Management System (ISMS) Documentation The policy and procedure set, plus the operational records that prove any of it happens. Information

ISO 27001 Business Continuity Plan

Two controls decide whether your ISO 27001 business continuity plan survives an audit: Annex A 5.29 and Annex A 5.30. One keeps your security controls working while everything else is failing. The other gets your systems back online before the damage becomes permanent. Plenty of teams write a continuity policy that satisfies neither in the way a certification auditor expects, and they discover the gap during the Stage 2 audit, when it is expensive to fix. This article covers what ISO 27001:2022 actually requires for business continuity, the components an auditor will ask to see, the step-by-step build, and the mistakes that turn a continuity plan into a non-conformity. What Is an ISO 27001 Business Continuity Plan? An ISO 27001 business continuity plan is the documented set of procedures that keeps information security effective and critical ICT services available during a disruption. It is not a generic “keep the lights on” binder. Under ISO 27001, the plan protects the confidentiality, integrity, and availability of information when normal operations break down: a ransomware event, a cloud outage, a data center failure, or a supplier collapse. The plan lives inside your Information Security Management System (ISMS). It draws on your risk assessment, your asset register, and your Business Impact Analysis (BIA), and it feeds your disaster recovery procedures. Scope is the part people get wrong. ISO 27001 cares about the information security aspects of continuity, not every operational hiccup a full business continuity program might cover.   Why You Need a Business Continuity Plan for ISO 27001 Compliance Downtime is expensive, and the bill arrives fast. For most organizations, the question is not whether a disruption will happen, but how quickly they recover when it does. There is also a hard compliance reason. You cannot certify to ISO 27001 while ignoring continuity. The standard requires you to maintain information security during disruption and to keep ICT able to support recovery, and an auditor will ask for the evidence. A continuity plan is where availability stops being a promise and becomes a tested capability. Let Axipro help you build a business continuity plan that’s practical, compliant, and audit-ready. Strengthen Your Business Continuity Strategy Schedule A Consultation ISO 27001 Requirements Related to Business Continuity Planning ISO/IEC 27001:2022 carries 93 Annex A controls across four categories: organizational, people, physical, and technological. Continuity sits in the organizational set, and two controls do the heavy lifting, supported by two more on the technical side. Annex A 5.29 – Information Security During Disruption A.5.29 requires you to maintain information security at an appropriate level when a disruption hits. The point is that security controls have a habit of degrading under pressure. People disable multi-factor authentication to “speed things up,” logging stops on a failover system, or access controls loosen while everyone scrambles. A.5.29 says the confidentiality and integrity of your information must be maintained even while availability is under threat. It is classed as both a preventive and a corrective control, meaning it should reduce the chance of an incident and also help resolve one already underway. Annex A 5.30 – ICT Readiness for Business Continuity A.5.30 is the technical engine. It requires that your ICT readiness is planned, implemented, maintained, and tested against business continuity objectives and ICT continuity requirements. In plain terms, your servers, networks, applications, and cloud services need a defined recovery path, each with a Recovery Time Objective (RTO) and Recovery Point Objective (RPO), and you need to prove the path works. This control is entirely new in the 2022 revision. It has no precedent in ISO 27001:2013, which is exactly why teams migrating from the older version so often have a gap here. Important: A.5.30 did not exist in ISO 27001:2013. If your continuity documentation was written against the old Annex A 17 cluster and never updated, you are missing a control the auditor will specifically test. Treat ICT readiness as a fresh requirement, not a relabel. Two technological controls back these up. Annex A 8.13 (Information Backup) requires backups to be taken and tested in line with an agreed policy, and Annex A 8.14 (Redundancy of Information Processing Facilities) covers the failover and redundancy that let critical systems keep running when a component dies. Relationship Between ISO 27001 and ISO 22301 This is where confusion is common. ISO 27001 requires the information security aspects of continuity. ISO 22301 is the dedicated standard for a full Business Continuity Management System (BCMS), covering people, facilities, supply chain, and operations far beyond information security. An ISO 27001 certificate does not certify your wider continuity program. The good news: both standards share the Annex SL high-level structure, so risk assessment, internal audit, management review, and document control carry across. Teams that already run ISO 27001 can layer ISO 22301 on top with far less effort than starting from scratch. Key Components of an ISO 27001 Business Continuity Plan Business Impact Analysis (BIA) The BIA is the foundation. It identifies your critical business processes, the ICT systems they depend on, and the cost of losing each one over time. It is where your recovery objectives come from, not from a vendor datasheet. A BIA also sets the Maximum Tolerable Period of Disruption (MTPD): the point beyond which an activity’s failure causes unacceptable damage. Risk and Disruption Scenario Assessment Your risk assessment identifies what could cause a disruption and how likely it is, feeding the Risk Treatment Plan and the Statement of Applicability (SoA) that records which controls apply. Continuity planning then runs concrete scenarios: ransomware, a regional outage, a key supplier failure, the loss of a data center. Response and Recovery Strategies For each critical system, you define how you will respond and recover: failover to a secondary site, restore from backup, or switch to a manual workaround. This links incident response to crisis management, the executive-level decision-making that kicks in when an incident escalates beyond a routine fix. Roles and Responsibilities Name real people, not departments. “IT will handle it” is the single most common

ISO 27001 Certification Cost

Most companies pursuing ISO 27001 certification cost analysis for the first time will spend between $10,000 and $50,000 in year one, and far less than half of that goes to the auditor. A 50-person SaaS company typically pays $10,000 to $22,000 in certification body fees alone, then doubles or triples that figure in implementation work, tooling, and internal hours before the Stage 2 audit even begins. The wide range exists because ISO 27001 certification cost is not a price tag; it is the sum of a dozen separate decisions: your scope, your security maturity, your certification body, and whether you build the ISMS yourself, hire a consultant, or run it through a compliance automation platform. This article breaks down every one of those costs, stage by stage and region by region, including the ones that never appear in vendor quotes. What Determines ISO 27001 Certification Cost? Six variables drive almost all of the variance between a $10,000 certification and a $150,000 one. Company Size and Employee Count Headcount is the single biggest cost driver because certification bodies calculate audit days (mandays) primarily based on the number of people working within the scope of your Information Security Management System (ISMS). The calculation is not arbitrary: accredited bodies follow the audit time tables in ISO/IEC 27006, which means a 20-person company and a 200-person company will receive structurally different quotes no matter how hard they negotiate. More employees also means more interviews, more evidence sampling, and more Annex A controls applied across more people. Scope and Complexity of the ISMS Scope is the variable you actually control. Your Statement of Scope defines which business units, systems, products, and locations fall inside the ISMS. A scope limited to one product line and the engineering team that runs it costs dramatically less to implement and audit than a whole-of-company scope. Complexity compounds this: bespoke infrastructure, regulated data types, and heavy third-party dependency chains all add controls, evidence, and audit time. Number of Physical and Cloud Locations Each physical site within scope can require its own audit visit, with travel costs on top. Multi-site organisations can reduce this through sampling (more on the square root rule later), but every additional location still adds something. Cloud environments count too: multiple cloud providers, regions, and tenancy models expand the technical scope auditors must cover, even when no travel is involved. Existing Security Maturity A company that already runs access reviews, maintains an asset inventory, and documents its incident response process is buying a much shorter journey than one starting from a blank page. The gap analysis exists precisely to price this difference. Organisations already aligned to SOC 2, NIST CSF, or Cyber Essentials Plus typically reuse 50 to 70 percent of their existing controls and evidence, which translates directly into lower implementation cost. Choice of Certification Body Certification bodies are not interchangeable on price. Large international names like BSI, Bureau Veritas, LRQA, and DNV charge premium day rates, often 30 to 50 percent above smaller accredited bodies, and their brand carries weight with enterprise procurement teams. What matters most is accreditation: a certificate issued by a body accredited by UKAS, ANAB, or another IAF (International Accreditation Forum) member carries international recognition. An unaccredited certificate is cheaper and close to worthless in serious sales conversations. Internal vs. External Implementation Approach The final driver is who does the work. Internal teams cost salary hours. Consultants cost fees. Platforms cost subscriptions. Each approach lands at a very different total, which is why this article dedicates a full section to it below. Average ISO 27001 Certification Cost Ranges The ranges below cover total first-year cost: implementation, tooling, and certification audits combined. They assume an accredited certification body and a sensibly defined scope. Cost for Small Businesses and Startups (1–50 Employees) A focused startup with a single product, cloud-native infrastructure, and a tight scope can realistically certify for $10,000 to $35,000 all-in. Lean implementations using templates or an automation platform sit at the bottom of that range. UK micro-businesses can find UKAS-accredited audit fees starting around £6,250, with day rates near £1,250. Cost for Mid-Sized Organizations (50–250 Employees) This is where most certifications happen, and where costs spread widest. Expect 8 to 12 initial audit days, $30,000 to $80,000 in total first-year spend, and a six to nine month timeline. Multiple departments, more mature customer requirements, and the first real multi-team coordination overhead all show up in the budget. Cost for Large Enterprises (250+ Employees) Enterprise certifications routinely exceed $100,000 in year one once you include program management, multiple sites, and large-scale audits. The audit fee alone can pass $50,000 for complex, multi-site scopes. At this scale, the internal time investment, covered under hidden costs below, often outweighs every external invoice. ISO 27001 Cost Breakdown by Stage Here is where the money actually goes, in roughly the order you will spend it. Cost of Purchasing the ISO 27001 Standard The official ISO/IEC 27001:2022 document costs CHF 155 (roughly $170) from the ISO store. Most teams also buy ISO 27002, the implementation guidance for the Annex A controls, for a similar amount. Budget $300 to $400 for both. Do not skip this purchase: implementing against second-hand summaries of the standard is a common source of audit findings. Gap Analysis Costs A consultant-led gap analysis before committing to anything else runs $2,000 to $10,000 depending on scope, while platform-based readiness assessments are often bundled into the subscription. The output, a clear map of where you stand against every clause and control, is what makes the rest of the budget predictable. ISMS Implementation Costs This is the largest and most variable line item: building the risk assessment, the risk treatment plan, the Statement of Applicability (SoA), and operationalizing the controls you have selected. Done internally, it consumes 200 to 600 hours of staff time over four to eight months. Done with consultants, expect $10,000 to $50,000 in fees for a typical SMB. Documentation and Policy Development Costs ISO 27001 requires a defined set of documented

Secure Data Disposal ISO 27001 SOC 2 GDPR

Researchers who buy second-hand drives off online marketplaces keep finding the same thing: live data.  A widely cited study by Blancco Technology Group found that 42% of used drives sold on eBay still held recoverable information, including financial records and personal data the previous owners assumed was long gone. The drives were not hacked; they were thrown away by organizations that treated deleting a file as the same thing as destroying it. Secure data disposal is where many compliance programs fail. ISO 27001, SOC 2, and GDPR all demand it, but they describe it in different languages, enforce it through different mechanisms, and punish failure in very different ways.  This article sets out what each framework requires, where the requirements overlap, and how to run a single disposal program that satisfies all three at once. Why Secure Data Disposal Matters Across Compliance Frameworks Disposal is the last link in the data lifecycle, and the easiest one to skip. An organization can run flawless access controls, encryption, and monitoring for years and still cause a reportable breach the moment one unwiped laptop leaves the building. A recoverable drive in a recycling skip is functionally identical to an open database on the internet, and auditors and regulators know it. Most disposal failures are unforced errors: a control that was already written into policy but never carried through to the actual hardware. The gap between having a disposal policy and proving this specific drive was destroyed is exactly where audits and breach investigations live. Defining Secure Data Disposal: Key Terms and Concepts What Is Secure Data Disposal? Secure data disposal is the end-to-end process of removing data and the equipment that holds it from active use, in a way that prevents its recovery. It covers the full lifecycle end: deletion of data while a system is still live, sanitisation of media that will be reused, physical destruction of media that will not, and the safe handling of equipment that is recycled, returned to a lessor, or sold. Disposal is the goal. The methods are how you get there. What Is Secure Data Destruction? Secure data destruction is the subset of disposal that renders media permanently unusable or its contents mathematically irretrievable. Shredding a drive, pulverising it, incinerating it, or destroying the encryption keys that make an encrypted disk readable are all forms of destruction. Destruction is one route to disposal, and it is the right route when the data is highly sensitive, or the media will never be reused. Secure Data Disposal vs. Secure Data Destruction: What Is the Difference? The distinction matters more than it looks. Disposal is the outcome you owe to every framework: data gone, unrecoverable, equipment handled appropriately. Destruction is just one of the methods. You can dispose of data without destroying the hardware by sanitising a drive thoroughly enough to reuse it. Confusing the two leads to two classic mistakes: destroying assets that could have been securely wiped and reused, and assuming a quick deletion counts as disposal when it does not. Important: Emptying the recycle bin, formatting a drive, or hitting delete does not dispose of data under any of these frameworks. Standard deletion only removes the pointer to the data; the bits remain until they are overwritten. Every framework discussed here expects the data to be unrecoverable, which is a far higher bar than not visible. What ISO 27001 Requires for Secure Data Disposal ISO/IEC 27001 handles disposal through a small cluster of Annex A controls that auditors read as a single process rather than in isolation. The two controls that do most of the work are 7.14 and 8.10. For a deeper look at how these controls fit into a broader compliance program, see our ISO 27001 implementation guide. ISO 27001 Annex A 7.14: Secure Disposal or Re-Use of Equipment Annex A 7.14 is a physical control. Before any equipment is disposed of or reused, the organisation must check whether it holds information assets or licensed software and ensure those are permanently erased or the media physically destroyed. It applies to servers, laptops, desktops, mobile devices, printers, network gear, and any storage media: if it ever processed information, it is in scope. The control replaces the older 2013 clause 11.2.7 and adds explicit expectations around removing identifying markings and handling end-of-occupancy scenarios. ISO 27001 Control 8.10: Information Deletion Annex A 8.10 is a technological control, and it focuses on the data rather than the box. It requires information stored in systems, devices, or media to be deleted when it is no longer required, and rendered unrecoverable. The cleanest way to keep these straight: 8.10 governs the data while it is in use or reaches its retention limit; 7.14 governs the hardware at end of life. Most retention-driven deletion sits under 8.10; most decommissioning sits under 7.14. ISO 27001 Control 8.12: Data Leakage Prevention and Its Role in Disposal Control 8.12 is rarely filed under disposal, but improperly discarded media is one of the oldest data leakage channels there is. A drive that leaves your control with recoverable data on it is a leak, regardless of how it left. Treating disposal as part of your leakage prevention posture forces the right question at the right time: what could walk out the door on this device, and has it actually been removed? Physical Destruction and Irretrievable Erasure Under ISO 27001 ISO 27001 offers two broad routes: physically destroy media that holds information, or erase and overwrite it so retrieval by a malicious party is precluded. The standard cross-references ISO/IEC 27040 for detailed sanitisation methods. The unifying requirement is that recovery should be impractical, not merely inconvenient. Deletion alone never satisfies this. Overwriting, Full-Disk Encryption, and Other Approved Methods Overwriting user-accessible storage with multiple passes is acceptable for many sensitivity levels. Full-disk encryption changes the economics of disposal entirely: if a device is encrypted from day one and the keys are properly managed, secure disposal can be as simple as destroying the keys, a technique known as

Business Continuity Plan Testing for SOC 2

A business continuity plan that has never been tested is, to a SOC 2 auditor, a document and nothing more. The Availability criteria do not award credit for a polished plan sitting in a shared drive. They ask for evidence that you ran the plan, watched it work or fail, recorded what happened, and fixed what broke. That gap — between having a plan and proving it works — is where most availability findings originate. Business continuity plan testing for SOC 2 is the exercise that turns your plan into auditable evidence. It maps directly to Availability criterion A1.3, one of the few SOC 2 controls that explicitly requires you to test something rather than merely document it. This guide covers what counts as a valid test, the test types auditors accept, a step-by-step process, the exact evidence you need, and the mistakes that turn a routine review into a finding. What Is Business Continuity Plan Testing in the Context of SOC 2? Business continuity plan (BCP) testing is the structured validation of whether your organization can keep critical operations running — and restore them within defined targets — during a disruption. In a SOC 2 context, the testing is not freeform. It must produce dated, traceable evidence that the recovery procedures in your plan actually work, that the people involved know their roles, and that systems and data come back within your stated recovery objectives.   Why SOC 2 Requires Business Continuity Plan Testing SOC 2 is an attestation against the AICPA’s Trust Services Criteria, and the Availability category exists specifically for organizations that make uptime or resilience commitments to customers. A plan you never exercise cannot demonstrate operating effectiveness over the audit period — which is the entire point of a Type 2 examination. Testing is the control that converts a static plan into a recurring, observable activity an auditor can sample. SOC 2 Trust Services Criteria and BCP Testing Requirements Availability is one of the five Trust Services Criteria, and it is optional, included only when your service commitments warrant it. When in scope, it is built around three sub-criteria: A1.1 addresses capacity management. A1.2 addresses recovery infrastructure and backup processes. A1.3 addresses the testing of recovery procedures. BCP testing lives squarely in A1.3, with A1.2 supplying the backups and infrastructure that the test validates. Availability Criteria A1.2 and A1.3 Explained Per the AICPA’s Trust Services Criteria, A1.2 requires the entity to design, implement, operate, and monitor environmental protections, recovery infrastructure, and data backup processes that meet its availability objectives. In plain terms: you need real backups, stored away from production, with recovery infrastructure ready to use. A1.3 then requires the entity to test recovery plan procedures supporting system recovery to meet its objectives. The two work as a pair: A1.2 builds the capability, A1.3 proves it functions. Important: The most common A1.3 gap is not a missing test. It is a test that never validated the recovery objectives. Teams run a tabletop, write “no issues found,” and move on — but the plan claims a 4-hour RTO that no one ever measured against an actual restore. If your plan states recovery targets, your test evidence must show whether you met them. A test that does not measure against your RTO and RPO leaves the most important question unanswered.   What Auditors Look for During a BCP Test Review Auditors want proof that the test happened, proof that it was meaningful, and proof that it led somewhere. Concretely, that means a test plan with a defined scenario, a dated record of execution with participants, results measured against your recovery objectives, a list of gaps or issues found, and evidence that those issues were remediated. A test that finds nothing and changes nothing is treated with suspicion — because real tests almost always surface something.   Types of Business Continuity Plan Tests Accepted for SOC 2 SOC 2 does not mandate a specific test type. It expects the rigor of the test to match the criticality of what you are protecting. The four common approaches sit on a spectrum from low-effort, low-disruption to high-effort, high-assurance. Tabletop Exercises A tabletop exercise is a facilitated discussion where key personnel talk through a disruption scenario and their responses. It is cheap, fast, and excellent for confirming that people understand their roles and that the plan reads coherently. Its limit is obvious: nobody actually recovers anything. For many organizations a tabletop is a legitimate annual test, especially in the first audit cycle, but auditors expect more rigor as a program matures. Walkthrough and Simulation Tests A simulation applies a specific scenario and asks the team to perform recovery actions, not just describe them. It is more involved than a tabletop and far better at exposing the gaps that only appear when people touch the tools. Simulations are where teams discover that a runbook references a system that was decommissioned, or that the on-call engineer lacks the access the plan assumes. Full Interruption Tests A full interruption test shuts down primary systems and shifts operations entirely to the recovery environment. It is the most comprehensive validation available and the only one that proves your failover genuinely works end to end. It also carries real operational risk, so it demands thorough planning and is usually reserved for mature programs and the most critical systems. Parallel Testing Parallel testing activates recovery systems alongside production without taking the primary offline, then compares the two to confirm the recovery environment performs as expected. It delivers much of the assurance of a full interruption test while sparing the business the disruption. For most SaaS and cloud-hosted services, parallel testing of failover and restore is the sweet spot between confidence and risk. How to Test Your Business Continuity Plan for SOC 2 Compliance The sequence below aligns with the contingency planning process in NIST’s Contingency Planning Guide, SP 800-34, which auditors widely treat as authoritative for resilience practices. Each step produces an artifact, and the artifacts together form

Employee Offboarding Checklist

31% of organizations have caught former employees accessing SaaS applications after their departure (source). Seventy percent of intellectual property theft happens in the ninety days surrounding a resignation announcement. The pattern is so consistent that auditors now treat termination day as one of the highest-risk windows on the security calendar. This article is a working employee offboarding checklist for IT, security, and HR teams who want to close that window cleanly. It walks through ten steps that revoke access without leaving gaps, then covers edge cases (remote workers, hostile exits, lost devices), the manual-versus-automation tradeoff, and post-offboarding monitoring. Use it as a baseline and adapt it to your environment. What Is Employee Offboarding and Why Does Access Revocation Matter? Employee offboarding is the structured process of separating a person from an organization: removing their access, recovering company property, documenting their exit, and updating records. The access revocation piece is the part where most programs fail quietly. Accounts get disabled in the identity provider but stay active in a dozen SaaS tools. Badges get collected but VPN tokens stay valid. The person is gone; the keys to the building are not. Why Employee Offboarding Is a Critical Security Risk Offboarding fails because access has multiplied faster than the processes designed to manage it. The average enterprise now operates somewhere between 275 and 660 SaaS applications depending on size, with employees touching dozens of them each week. Each application is a separate place that needs to be cleaned up, and each one creates an independent point of failure. The departing employee is a particularly acute version of this risk because the motivation to walk away with something often peaks during the same window that access is supposed to be revoked. The Cost of Leaving Access Open After Departure The financial picture is well documented. The 2025 Ponemon Cost of Insider Risks report puts the average annual cost of insider-related incidents at $17.4 million per organization, with containment taking an average of 81 days. Even when a departed employee never actively misuses their access, the existence of a forgotten account is enough to compromise a SOC 2 audit, trigger a breach notification, or create the credentialed beachhead that an outside attacker eventually exploits. The cases keep appearing. Cash App was breached in 2022 when a former employee accessed the records of 8 million customers after leaving. In May 2024, FinWise Bank disclosed that a former employee accessed internal systems after departure because access had never been fully revoked. Intel sued a former engineer in 2024 for downloading roughly 18,000 sensitive files in the days before he left. Ponemon’s 2025 report found that containment costs scale steeply with time. Incidents resolved in under 30 days averaged about $11 million, while those over 90 days averaged $17 million. The biggest variable is not detection capability. It is how fast access actually came down on day one. Compliance and Legal Implications of Incomplete Offboarding Access revocation is not a “best practice.” It is an explicit control requirement in nearly every framework against which an organization is likely to be audited. NIST SP 800-53 control PS-4 requires that on termination, organizations disable system access within an organization-defined time period, terminate or revoke any authenticators, and retrieve organizational property. ISO/IEC 27001 includes equivalent expectations under its Annex A controls for termination of employment. The AICPA Trust Services Criteria for SOC 2 cover this under Common Criteria CC6.2 and CC6.3, and auditors routinely pull a sample of terminated employees and verify timestamps in the identity provider against the HR system. GDPR adds a separate dimension. If a former employee still has access to the personal data of EU residents, that constitutes unauthorised processing under Article 32, and it is the controller’s responsibility, regardless of intent. HIPAA does the same for protected health information. Whatever the framework, the question an auditor or regulator will ask is the same: how quickly was access revoked, and can you prove it? Who Is Responsible for Employee Offboarding? Offboarding fails most often because no one owns the whole process. Four groups need to be in the loop, and each one has a distinct job. HR and People Operations HR is the source of truth for the termination event. Their job is to capture notice of departure, set the official last day, communicate timing to the rest of the business, and serve as the trigger that starts every downstream task. If HR does not record the termination in the HRIS, nothing automated will fire. IT and Security Teams IT executes the access teardown. They disable accounts in the identity provider, revoke SSO and OAuth tokens, remove SaaS application access, suspend email, and recover devices. Security teams typically run the audit trail and post-offboarding monitoring, and they are the ones answering when an account flagged six months later turns out to belong to a person who left in March. Legal and Compliance Legal handles NDA reminders, IP assignment confirmations, non-disclosure obligations, and any contractual surprises. Compliance owns the documentation: the evidence trail that proves the offboarding actually happened and met the relevant control requirements. For regulated industries this becomes audit evidence; for everyone else it becomes legal cover. Direct Managers Managers know things HR does not. They know which shared drives the person owned, which third-party vendors they had standing access to, which client passwords they may have rotated themselves, and which projects need a transition plan. A solid offboarding process forces the manager into the workflow with a checklist of role-specific items, because no central team can guess them. Employee Offboarding Checklist: 10 Steps to Revoke Access Without Leaving Gaps This is the core sequence. The order matters: starting with notification and inventory before disabling accounts means you do not lock the person out of a system you still need them to hand off. Step 1: Initiate Offboarding Immediately Upon Notice of Departure The moment notice is given — resignation, termination decision, or end of contract — the offboarding workflow should start. This means

ISO 27001 and GDPR

Plenty of companies treat an ISO 27001 certificate as proof of GDPR compliance. It is not. The two frameworks overlap heavily, but they answer different questions, and the gap between them is exactly where regulators tend to look. ISO 27001 tells you how to build a defensible security program. GDPR tells you what the law expects when that program touches personal data. Run one without understanding the other, and you will either over-engineer security you do not strictly need, or miss privacy obligations that carry real financial exposure. This article maps where ISO 27001 and GDPR meet, where they part ways, and how to run them as a single coordinated effort rather than two competing projects. What Is ISO 27001? ISO/IEC 27001 is the international standard for an Information Security Management System, or ISMS. The current edition is ISO 27001:2022. It is not a checklist of technical fixes. It is a management framework: a structured, repeatable way to identify information security risks, decide how to treat them, document those decisions, and improve over time. Clauses 4 to 10 of the standard define the mandatory ISMS requirements, covering leadership, risk assessment, internal audit, and management review. Annex A then lists 93 controls grouped into four themes: organisational, people, physical, and technological. You do not implement all 93 by default. You select the controls that address your assessed risks and justify your choices in a document called the Statement of Applicability. Certification against ISO 27001 is voluntary and is granted by an accredited third-party body after an audit. What Is GDPR? The General Data Protection Regulation is European Union law. It has been applied since 25 May 2018, and it applies to any organisation that processes the personal data of people in the EU, wherever that organisation is based. GDPR is fundamentally about the rights of individuals, not just the security of data. It grants people rights over their personal data, including access, correction, erasure and portability. It places obligations on the organisations that decide how data is used (controllers) and those that process it on their behalf (processors). It requires a lawful basis for every processing activity, mandates breach notification, and demands transparency about what happens to people’s information. You do not implement GDPR and receive a certificate. You obey it, and a regulator decides whether you have. Key Differences Between ISO 27001 and GDPR Scope and Purpose ISO 27001 protects all information assets an organisation holds: intellectual property, financial records, operational data, source code and, yes, personal data. Its purpose is the confidentiality, integrity and availability of information in general. GDPR is narrower in one sense and broader in another. It covers only personal data of individuals in the EU, but it protects the person behind the data, not merely the data itself. A system can be flawlessly secure and still violate GDPR. Legal Obligation vs. Voluntary Certification This is the difference that catches people out. GDPR is binding law. If you process EU personal data, compliance is not optional, and there is no opting out. ISO 27001 is a voluntary standard. Organisations pursue it for assurance, for competitive advantage, and because customers increasingly demand it. Crucially, there is no such thing as a GDPR certificate. Regulators assess compliance through investigation and enforcement, not through a badge you can display. Penalties for Non-Compliance GDPR fines run on two tiers under Article 83. Less severe infringements — such as failures around records of processing or breach notification — can reach €10 million or 2% of global annual turnover, whichever is higher. The more serious tier, covering breaches of the core processing principles and data subject rights, can reach €20 million or 4% of global annual turnover. Failing an ISO 27001 audit carries no legal fine at all. The consequence is commercial: you do not get the certificate, or you lose it, and that can cost you contracts. How ISO 27001 and GDPR Align Despite their different purposes, the two frameworks were built on compatible logic, which is why running them together works. Both treat information security as central. GDPR Article 32 requires “appropriate technical and organisational measures” to secure personal data. That phrasing is almost a direct description of what an ISO 27001 ISMS produces. The controls an organisation selects for confidentiality and access already serve the regulation’s security expectations. Both are risk-based. ISO 27001 starts every control decision from a risk assessment. GDPR expects the same proportionality: the measures you apply should match the sensitivity of the data and the likelihood and severity of harm. One risk methodology can serve both, provided you assess personal data processing risks alongside broader security risks. Both demand incident response. ISO 27001’s incident management controls require organisations to detect, assess and respond to security events. GDPR Article 33 requires notifying the supervisory authority of a personal data breach within 72 hours of becoming aware of it. The ISO process is the engine that makes the GDPR deadline achievable. How ISO 27001 Can Help You Comply With GDPR Four areas of an ISMS do direct, practical work toward GDPR compliance. Asset management. ISO 27001 requires an inventory of information and associated assets, with owners assigned. You cannot protect personal data, respond to access requests, or maintain records of processing if you do not know where that data lives. The asset inventory is the foundation for both frameworks. Access control. Identity management, privileged access controls and the principle of least privilege limit who can see personal data. That directly supports the GDPR requirement to ensure confidentiality and to prevent unauthorised access. Operational security. Logging, malware protection, backup and secure configuration keep personal data accurate, available and resistant to compromise. These map cleanly onto the integrity and availability expectations in Article 32. Techniques such as data masking for GDPR and ISO 27001 also sit within this space, reducing exposure without sacrificing operational utility. Incident management. A defined process for detecting and handling security events gives you the evidence trail and the response capability you need to

AI Hub

The latest insights into AI implementation and compliance

OWASP published the 2026 edition of its Top 10 for LLM Applications on August 4, 2026, during Black Hat week, and eight of the ten entries changed position. One got renamed. The message behind the reshuffle is blunt: you won’t build a model that can’t be fooled, so build the application around it in a way that limits the damage when it is. That one idea explains almost every move in the new ranking, and it should change how your team thinks about shipping AI features. This guide walks through the 2026 list in plain English: what each risk means, a real-world example, and what your team can actually do about it, with or without a dedicated security function. What Is the OWASP GenAI LLM Top 10 2026? The OWASP Top 10 for LLM Applications is a community-built awareness document that ranks the ten most critical security risks in applications powered by large language models. The OWASP GenAI Security Project, a global open-source initiative under the OWASP Foundation, maintains it, and the 2026 edition is the third release since the list first appeared in 2023. OWASP, the Open Worldwide Application Security Project, has published risk lists for web applications since 2003, and those lists became the shared vocabulary security teams, auditors, and buyers use to talk about risk. The GenAI LLM Top 10 does the same job for AI. Whether you’re a two-person startup wiring an API into a chatbot or an enterprise running retrieval pipelines, it gives you a common map of what actually goes wrong. One scoping note matters before anything else. The 2026 edition covers the model as a component inside an application: something that accepts input, generates output, and maybe retrieves information. The moment the model becomes an actor, with tools it can call and consequences it sets in motion, the risk shifts to the companion OWASP Top 10 for Agentic Applications from December 2025. Most products now do both, so most teams need both lists. Why the 2026 Update Matters for AI Builders Two things separate this edition from everything OWASP has published on AI so far. First, the methodology changed. Every previous version rested purely on expert consensus, meaning hundreds of practitioners voting on which risks matter most. This time the vote carried 75% of the weight, and the remaining 25% came from analysis of 6,639 real-world AI security incidents pulled from public vulnerability databases and an AI-harm database. It’s the first edition grounded in evidence of what has actually gone wrong rather than expert prediction of what might. Second, the framing changed. The project leads open the 2026 release by telling teams to stop optimizing the model and start optimizing the containment. The industry has spent two years pouring effort into filters, guardrail models, and jailbreak resistance. The 2026 list says: assume those will eventually fail, and make sure that when they do, nothing important breaks. AI security becomes blast radius control rather than perfect prevention. And this isn’t just a security engineer’s document. Developers decide what tools and permissions a model gets. Product owners decide which workflows run without a human in the loop. Founders and ops leads are the ones answering the security questionnaires where these questions now show up. The 2026 edition also ships a mapping appendix that connects every risk to frameworks your customers and auditors already recognize: NIST’s AI Risk Management Framework, MITRE ATLAS, MITRE CWE, and the Agentic Top 10. Insider Note: Enterprise vendor assessments have started asking about the OWASP LLM Top 10 by name. In security questionnaires we complete for clients at Axipro, questions like “describe your controls against prompt injection and excessive agency” began appearing in early 2026, sometimes before the buyer’s own team could explain what they meant. Being able to answer with a mapped control set is becoming a deal-cycle advantage, not just a security exercise. How the 2026 List Differs From Previous Versions The top two entries held their positions. Everything below them moved. Key Shifts Since the 2025 Update Excessive Agency jumped from sixth to third, the biggest promotion on the list. In 2025, giving a model tools and autonomy was mostly a theoretical worry. By 2026, agentic deployments had produced real production incidents, and the community concluded that agency is what decides whether a successful prompt injection is an inconvenience or a breach. Unbounded Consumption rose four places, from tenth to sixth. Inference costs became a real budget line as reasoning models, long outputs, and agent loops multiplied the compute behind a single request. “Denial of Wallet,” where an attacker spends pennies to trigger spend you can’t afford, is now a mainstream finding. Improper Output Handling fell from fifth to tenth. The risk didn’t shrink. It fell because it’s well understood and directly fixable with encoding and validation practices web developers already have. The entries above it are neither. What’s New, Renamed, or Reprioritized System Prompt Leakage became Hidden Context Exposure, and the scope widened a lot. The 2025 entry worried about attackers extracting your system prompt. The 2026 entry covers everything assembled into the model’s context that users aren’t meant to see: system instructions, retrieved policy documents, tool schemas, workflow rules. The guidance is unusually honest for a security document: assume all of it is discoverable, and design so that disclosure costs you nothing. Data and Model Poisoning absorbed fine-tuning subversion. The attack surface for corrupting a model’s behavior runs from pretraining data through fine-tuning pipelines into the retrieval stores RAG systems depend on, and the entry now says so. Misinformation climbed on evidence, not opinion. Practitioners voted it low; the incident data ranked it high. As reported in Help Net Security’s coverage of the release, OWASP also describes a “defense effect” working in the opposite direction on prompt injection: teams block it so effectively that few successful attacks reach public databases, which makes the risk look smaller than the money spent containing it. Signals About Where AI Security Is Heading Read together, the moves point one

For the past two years, enterprise AI risk conversations have centered on a familiar set of concerns: model bias, hallucination, data privacy, and dependency on third-party models. These are real risks, and most organizations now run some version of a governance program to manage them. But something has shifted. Organizations are no longer just deploying AI that generates content for a human to review. They’re deploying AI that acts. Agents now plan multi-step tasks, call APIs, move data between systems, execute transactions, and coordinate with other agents, often with no human checkpoint in the loop. That shift deserves more than a footnote in the existing AI risk category. It deserves its own line in the risk register: Agentic Autonomy Risk. What Is Agentic AI Risk Management? Agentic AI risk management is the practice of identifying, assessing, and controlling the risks created when AI systems take autonomous action on an organization’s behalf. Where traditional AI governance evaluates outputs (accuracy, bias, privacy), agentic AI risk management governs what agents actually do: the tools they call, the permissions they inherit, and the downstream consequences of their actions. That distinction is the reason existing risk registers struggle with agents, and it’s worth unpacking properly. What Agentic AI Actually Changes Traditional AI systems, even generative ones, are advisory. They produce an output such as a summary, a prediction, a draft email, or a classification, and a human remains the last checkpoint before anything happens in the real world. Agentic AI removes that checkpoint. An agentic system doesn’t just produce an answer. It pursues a goal. It decides which tools to call and in what order, then executes those actions directly against live systems: submitting a purchase order, modifying a database record, sending an external communication, or orchestrating a set of sub-agents to complete a broader workflow. Agentic autonomy is the degree to which a system can plan and execute actions without a human explicitly authorizing each step. It’s a spectrum rather than a binary. At one end, the AI drafts and a human approves every action. At the other, the AI operates within broad guardrails and only escalates exceptions. The further an organization moves along that spectrum, the less its exposure looks like software risk and the more it looks like delegated authority risk, the kind normally reserved for employees, contractors, and automated financial systems. Why Existing Risk Registers Miss Agentic AI Risks Most enterprise risk registers were built on a reasonably safe assumption: a human initiates consequential actions, and the technology around that human behaves deterministically. Agentic AI breaks both halves of that assumption at once. A few specific gaps show up quickly when organizations try to map agentic deployments onto existing categories. Operational risk registers assume process failures come from human error or system outages, not from a system independently choosing an unanticipated path to a stated goal. Cybersecurity risk registers are built around unauthorized external access, while an agent problem usually involves an authorized system taking unauthorized internal actions with its own legitimate credentials. Model risk frameworks, borrowed largely from financial services, evaluate output accuracy rather than action consequences, which matters most when those actions can’t be reversed. And third-party risk assessments treat vendors as static entities, not as autonomous agents that might invoke other vendors’ agents on your behalf. See our guide to the NIST AI Risk Management Framework for how output-focused frameworks are structured. The result is a governance blind spot. An organization can be compliant against its AI policy, its cybersecurity policy, and its vendor risk policy, and still have nobody accountable for the specific risk of a system initiating a harmful sequence of actions before anyone notices. Defining Agentic Autonomy Risk Agentic Autonomy Risk is the risk that an AI system, operating with delegated decision-making and execution authority, takes actions that are harmful, non-compliant, or misaligned with organizational intent before adequate human oversight can intervene. Those actions might happen independently or in coordination with other agents. It deserves standing as a named category alongside cybersecurity, operational, legal, financial, and third-party risk because the loss event itself is different. The harm is a completed action in a live system, and it may be difficult or impossible to reverse. The accountability structure is different too: when an orchestrating agent delegates to sub-agents, responsibility for the outcome gets distributed in ways existing ownership models don’t cleanly capture. So is the detection window. Traditional controls assume a human is positioned to catch an error before it compounds, but an agent can execute dozens of dependent actions faster than any human review cycle. 7 Agentic AI Risk Scenarios to Put on Your Register 1. Unauthorized autonomous decision-making. An agent takes an action within its technical permissions but outside its intended business mandate. It adjusts pricing, approves a refund, or modifies a customer record, and no policy ever explicitly authorized that scenario. 2. Goal misalignment. The agent optimizes for a literal interpretation of its objective in a way that diverges from actual business intent, particularly under ambiguous or adversarial inputs. 3. Multi-agent interactions and cascading failures. One agent’s flawed output becomes another agent’s trusted input. A single error can propagate across a chain of agents faster than anyone can detect it, amplifying the original mistake instead of containing it. 4. Excessive tool or system permissions. Agents get provisioned with broad, standing access “to be safe” rather than scoped, least-privilege access tied to specific tasks. A productivity tool quietly becomes a privilege-escalation path. 5. Regulatory non-compliance. Autonomous actions trigger obligations under data protection, financial services, employment, or sector-specific regulation, and they execute without the compliance review a human-initiated process would normally receive. 6. Explainability and accountability gaps. An autonomous action causes harm and the organization can’t clearly reconstruct why the agent chose that path, or establish whether the business owner, the AI governance function, or the vendor is accountable for the outcome. 7. Autonomous third-party actions. A vendor’s agent, integrated into your environment, takes action on your behalf, or your agent acts against a

Two compromised versions of LiteLLM sat on PyPI for roughly 40 minutes on the morning of March 24, 2026. That window was enough to capture secrets from around 434,000 CI/CD pipeline runs across nearly 2,500 organizations, including AWS, Samsung, Cisco, Salesforce, Siemens, and Deloitte. In August, researchers at CloudSEK and Hudson Rock confirmed they had obtained the raw exfiltrated data: a 153GB archive containing 433,909 files of environment variables, cloud keys, Kubernetes secrets, and API tokens harvested live from running pipelines, as covered by Help Net Security’s reporting on the credential archive. If LiteLLM runs anywhere in your stack, or you touch any AI proxy infrastructure at all, you need answers to three things: whether you were exposed, what to rotate first, and whether the rotation you did back in March actually held. That last one matters more than it sounds, because “we rotated everything” has already burned at least one very large company. How the Breach Happened The attack didn’t start with LiteLLM. On March 19, 2026, a threat group called TeamPCP compromised the build pipeline of Trivy, a vulnerability scanner half the industry runs, and pushed a poisoned release. LiteLLM’s own CI pipeline ran Trivy, so the poisoned scanner had legitimate read access to the project’s runner environment. The attackers used that to steal LiteLLM’s PyPI publishing tokens and ship two malicious releases of their own: versions 1.82.7 and 1.82.8. KICS and the Telnyx Python SDK got hit in the same campaign. The payload design is the part worth studying. The malicious package dropped a .pth startup hook into site-packages, so the code ran the moment any Python interpreter started on the machine, whether or not anything imported LiteLLM. From there it harvested environment variables, read local credential files like .aws/credentials and .kube/config, tried to move laterally across Kubernetes clusters, and installed a systemd backdoor dressed up as a generic telemetry service. InfoQ’s coverage of the PyPI compromise put downloads of the compromised release above 40,000. For scale, LiteLLM normally gets downloaded around 3 million times a day. The exfiltration had a nasty fallback, too. According to CloudSEK, stolen data was encrypted and sent to a typosquatted domain, and when that failed, the malware created a public repository inside the victim’s own GitHub account and uploaded the loot as a release asset. Some companies were publishing their own secrets to the open internet and had no idea. Worth Knowing: The malicious code only existed in the PyPI artifacts. The GitHub source repository stayed clean the whole time, so a developer reviewing the code on GitHub saw nothing wrong. Source review isn’t artifact verification. If you don’t check that what the registry serves matches the upstream source, this class of attack is invisible to you. How to Check If You Were Exposed Three checks, from quickest to most involved. 1. Confirm whether the compromised versions ever ran The malicious versions went live on PyPI at 10:39 UTC on March 24, 2026 and got quarantined about 40 minutes later. The project’s advice: treat any install from that day before 16:00 UTC as suspect. Search your lockfiles, pip caches, SBOMs, and container image histories for 1.82.7 and 1.82.8. And check your internal artifact mirrors. An Artifactory or Nexus proxy that cached the bad release in March can keep serving it internally long after PyPI pulled it. Keep the .pth mechanism in mind when you scope this. The question isn’t “which applications import LiteLLM,” it’s “which machines had the package installed at all,” because every Python process on an infected machine triggered the payload. 2. Hunt for persistence Rotation is pointless if the attacker still has a foothold. Check developer machines, CI runners, and containers for unauthorized .pth files in site-packages and for suspicious systemd units, especially anything posing as a system telemetry service. And review activity from March 24 onward, not just the 40-minute window. Persistence is there so the access outlives the infection. Pro Tip: Don’t limit the persistence hunt to live machines. Base container images rebuilt in late March may have baked the payload into every image derived from them since. Scan your image registry for the affected LiteLLM versions and for unexpected .pth files, then trace which running workloads came from flagged images. 3. Check whether your secrets are in the dump Hudson Rock has published a domain lookup tool and is running ethical disclosures for affected organizations, and CloudSEK maintains a high-confidence victim list. Use them, but know their limits. Attribution in this dataset is genuinely hard. One dump with a siriusxm.com committer email actually traced, through its self-hosted GitLab endpoints, to AdsWizz, a SiriusXM subsidiary. And a large share of the dumps are generic pipeline configurations with no identifying domain, email, or server name at all. Absence from a victim list is not evidence of absence. If your pipelines ran the compromised versions, assume exposure no matter what a lookup tool tells you. What to Rotate, in What Order The guidance from both research teams is blunt: treat every secret the LiteLLM environment could reach as compromised. That covers secrets on disk, in memory, injected into CI jobs, and anything retrievable through instance metadata services. Work down by blast radius: Priority Credential type Why it comes first 1 Cloud IAM keys (AWS, GCP, Azure) Direct control of infrastructure, data stores, and billing. This is where attackers monetize fastest. 2 GitHub and GitLab PATs, package publishing tokens These let an attacker poison your releases and turn your company into the next link in the supply chain. 3 Kubernetes service account tokens and kubeconfigs Lateral movement across clusters was built into the payload, not a theoretical risk. 4 Database passwords and third-party API keys Dumped in plain text in the archive, often with no attribution, so nobody will warn you they leaked. 5 AI provider API keys Billing abuse, quota theft, and access to whatever data flows through your LLM routing layer. One word matters more than the rest of this article: revoke, don’t just rotate. That

AI Tool Usage Tracking

One in five breached organizations last year traced the incident to shadow AI, and those breaches cost an average of $670,000 more than standard incidents, according to IBM’s 2025 Cost of a Data Breach Report. The worst part is that most of those organizations already ran a CASB, a DLP program, or both. The tools were on, but the traffic still got through. That’s the visibility gap this article is about. AI tool usage tracking isn’t the same problem as SaaS discovery, and the security stack built for the SaaS era misses most of what matters about AI. Below, we break down what tracking actually requires, where CASB and DLP fail, which categories of AI usage slip through, and what a stack that works looks like in 2026. What AI Tool Usage Tracking Actually Means Most teams that say they “track AI usage” mean they can see that someone visited chat.openai.com. That is app discovery, and it answers almost none of the questions a security or governance team actually needs answered. Beyond App Discovery: Tracking Prompts, Data Flows, and Model Interactions Real tracking covers three layers. First, which tools are in use: chatbots, copilots, coding assistants, embedded SaaS features, agents. Second, what data moves: the content of prompts, uploaded files, and pasted context, mapped against data classifications. Third, how models behave in your environment: which endpoints get called, which OAuth grants exist, which agents hold standing permissions. Seeing that an employee opened ChatGPT gets you nowhere. What you actually need to know is whether they pasted a customer contract into a personal account while they were there. The Difference Between Detection, Monitoring, and Continuous Tracking Detection is a point-in-time answer to “what AI is here?” Monitoring watches known tools on an ongoing basis. Continuous tracking is broader: it assumes the inventory changes weekly, correlates identity, data, and endpoint signals over time, and feeds a governance program rather than a one-off report. Frameworks such as the NIST AI Risk Management Framework and ISO 42001 assume the third mode. A discovery scan from last quarter won’t satisfy an auditor, and it certainly won’t slow down an attacker. Why Traditional SaaS Monitoring Falls Short for AI SaaS monitoring was built around a stable premise: an app is a destination with a domain, a login, and an admin console. AI breaks that premise in several ways at once. The risky activity is the content of an interaction, not the visit. The tool often isn’t a destination at all but a feature inside an app you already sanctioned. And increasingly the “user” isn’t a person but an agent acting on delegated credentials. Why CASB Misses Shadow AI Usage The Cloud Access Security Broker sits between users and cloud services to enforce policy, and for classic SaaS governance it still earns its keep. AI has structural blind spots that no amount of tuning can fix. CASBs Were Built for SaaS Apps, Not Model Endpoints A CASB catalog maps domains to applications with risk scores. AI usage doesn’t resolve neatly to a domain. The same api.openai.com endpoint serves a sanctioned enterprise deployment, a developer’s weekend experiment, and a data-leaking browser extension, and the catalog sees one “app”. Meanwhile, new model endpoints, wrappers, and niche AI tools appear faster than any vendor catalog can keep up with. Gartner research from late 2025 found 69% of organizations already suspect or have evidence that employees use prohibited public generative AI tools, catalog or no catalog. Blind Spots in Encrypted API Traffic to LLM Providers Prompt content travels over TLS. Without full TLS inspection, a CASB sees connection metadata: destination, volume, timing. It can’t see that the payload contained source code or patient records. And full TLS inspection is harder than the datasheet implies. Certificate pinning breaks it for many native apps and CLI tools, legal and works-council constraints limit it in the EU, and most organizations carve out broad exemption lists that AI traffic happily rides through. The OAuth and Embedded AI Problem CASBs Can’t See When an employee grants an AI meeting-notes tool access to their calendar and mailbox via OAuth, no proxy is involved at all. The vendor’s servers communicate directly with Microsoft’s or Google’s APIs using a persistent token. The same applies to AI features embedded inside sanctioned SaaS, think Notion AI, Slack AI, or Salesforce Einstein. The CASB sees approved traffic to an approved app, while the AI processing happening inside it, and whichever sub-processor it forwards data to, stays invisible. Personal Accounts and BYO-AI Bypass CASB Proxies Netskope’s 2026 Cloud and Threat Report found that nearly half of employees who use generative AI at work do so through personal accounts. Personal accounts on managed devices are hard enough; personal accounts on personal devices, home networks, and mobile connections never touch the corporate proxy path at all. Tenant restrictions help for a handful of major providers and do nothing for the long tail. Browser-Based and Extension-Delivered AI Escape Network Inspection AI browser extensions read page content and form inputs locally, then exfiltrate via their own backend, often to generic cloud infrastructure that categorizes as “technology” rather than “AI”. From the network’s view, it is routine HTTPS to a CDN. The riskiest interaction, an extension scraping everything an employee views, produces the most boring traffic signature. Insider Note: In AI governance readiness assessments, the OAuth grant review is where clients get the biggest surprise. We routinely find dozens of AI tools holding live mail, calendar, or drive scopes that nobody in IT ever approved, granted by employees who abandoned the tool (and sometimes the company) months earlier. The tokens keep working anyway. Why DLP Fails to Catch Shadow AI Data Exposure DLP has the opposite problem. It can sometimes see content, but it doesn’t understand it, and AI interactions defeat the pattern matching it depends on. Prompt-Based Data Loss Doesn’t Match DLP Signature Patterns DLP fires on signatures: credit card regexes, SSN formats, keyword dictionaries, file fingerprints. Sensitive prompts rarely look like that. “Summarize why we’re losing

Most organizations think their AI governance is further along than it is. McKinsey’s 2026 AI Trust Maturity Survey of roughly 500 organizations found an average maturity score of 2.3 out of 4, and only about a third reported level three or higher in strategy, governance, and agentic AI oversight. Adoption is outpacing control, and regulators have noticed. An AI governance maturity model gives you a way to measure that gap honestly. This guide covers what a maturity model is, the six dimensions it should measure, the five levels most models use, and how to assess your own organization and build a roadmap to the next level. What Is an AI Governance Maturity Model? An AI governance maturity model is a structured framework that describes how capable an organization is at governing its AI systems, usually across five progressive levels. The concept borrows directly from the Capability Maturity Model (CMM) that software engineering has used since the early 1990s: define the capability, describe what it looks like at each stage of development, and score yourself against it. The purpose is diagnosis. A maturity model tells you where governance is strong, where it’s theater, and where it doesn’t exist at all. How It Differs from General AI Governance Frameworks Frameworks like the NIST AI Risk Management Framework or ISO/IEC 42001 tell you what good governance contains: policies, risk assessments, accountability structures, monitoring. A maturity model tells you how well you’re doing those things today. The framework is the destination. The maturity model is the odometer. That distinction matters in practice. Plenty of companies can point to an AI policy document. Far fewer can show that the policy changes what teams actually ship. Why Enterprises Need a Maturity Model Three reasons. First, budget: you can’t prioritize governance investment without knowing which dimension lags. Second, accountability: a maturity score gives boards something concrete to track quarter over quarter. Third, regulation: the EU AI Act and frameworks like ISO 42001 assume a functioning management system, and a maturity assessment is the fastest way to find out whether yours would survive scrutiny. Core Dimensions of an AI Governance Maturity Model A useful model measures more than policy coverage. Six dimensions show up consistently across the credible models, including the IEEE-USA flexible maturity model built on the NIST AI RMF. Strategy and leadership. Does the organization have a stated position on AI risk, an executive owner (increasingly a Chief AI Officer), and board visibility? Gartner’s 2025 polling found 55% of organizations now have an AI board or dedicated oversight committee, which means nearly half still govern by improvisation. Policies, standards, and accountability. Written policies mapped to regulations, a RACI matrix for AI decisions, and clear escalation paths. Many organizations adapt the three lines of defense model from financial risk: the teams building AI, the risk function overseeing them, and internal audit checking both. Data governance and model lifecycle. Training data lineage, quality controls, and lifecycle management from development through deployment, monitoring, and retirement. This is where AI governance meets MLOps, and where mature organizations maintain an AI register, a live inventory of every model and system in production. Risk, compliance, and ethics. Risk classification of AI systems, impact assessments, bias and fairness testing, and explainability requirements. Banks will recognize the DNA of model risk management under SR 11-7 here. People, skills, and culture. Training, role clarity, and whether people outside the governance team actually understand their obligations. Tools, automation, and monitoring. Drift detection, automated policy checks, audit logging, and dashboards. Governance that lives in spreadsheets caps out around level three. The 5 Levels of AI Governance Maturity Level 1: Ad Hoc / Initial AI use happens without oversight. There’s no inventory, no policy, or a policy nobody follows. Shadow AI is common, and risk surfaces only when something breaks publicly. Level 2: Developing / Repeatable Someone has been assigned responsibility. A draft policy exists, a partial inventory exists, and reviews happen for high-profile projects. The practices are repeatable but depend on specific people rather than defined processes. Level 3: Defined / Structured Governance is documented, standardized, and applied across the organization. There’s a governance committee, a risk classification scheme, defined lifecycle gates, and mandatory training. Most organizations pursuing ISO 42001 certification are working to reach and formalize this level. Level 4: Managed / Metrics-Driven Governance produces numbers. Coverage rates, review cycle times, incident counts, and risk reduction are measured and reported to leadership. Controls are enforced by tooling rather than goodwill, and audits confirm the system works as described. Level 5: Optimized / Adaptive Governance improves itself. Monitoring feeds back into policy, controls adapt to new model types (agentic systems being the current test), and the organization anticipates regulatory change rather than reacting to it. Almost nobody is here yet, and that’s fine. Level 5 is a direction, not a deadline. Insider Note: In assessments, the most common self-scoring error is claiming level 3 on the strength of documents alone. If your policy says every model gets a pre-deployment review and your inventory shows 40 models but your review log shows 6, you’re at level 2. Evidence beats paperwork every time, and auditors check the logs first. AI Governance Maturity Matrix The matrix crosses dimensions with levels so you can score each one independently. Organizations are rarely uniform: it’s normal to sit at level 3 on policy and level 1 on monitoring. For scoring, keep the rubric simple: 1 to 5 per dimension, scored on evidence you could show an auditor, not on intentions. Board-level indicators (does the board see AI risk reporting?) and operational indicators (does every production model have a completed impact assessment?) should be scored separately, because they fail independently. How to Assess Your Current AI Governance Maturity Start with a baseline self-assessment. Pull together a cross-functional group covering engineering, legal, risk, security, and the business owners of major AI use cases, and score each dimension against the matrix. Half a day is usually enough for a first pass. For each dimension, the

Secure AI Agent Vendor

An AI agent that can read your inbox, query your CRM, and dig through internal documents has more standing access than most of your employees. It handles sensitive data, acts on its own, and often passes that data through sub-processors you’ll never see. Certifications are the quickest way to tell which vendors have let an outsider check their work, and which ones just put the word “secure” on a landing page. No single certificate proves an AI agent is safe. But the right mix of security attestations, privacy certifications, and AI governance standards tells you the vendor has real controls, that an independent auditor has tested them, and that someone is on the hook when the agent misbehaves. This guide covers which certifications to ask for, how to verify them, and which claims should make you walk away. The Core Certifications Every Secure AI Agent Vendor Should Hold SOC 2 Type II SOC 2 Type II is the baseline for any SaaS or AI vendor that handles customer data. A licensed CPA firm audits the vendor against the AICPA’s Trust Services Criteria (Security, Availability, Processing Integrity, Confidentiality, and Privacy) and reports on whether its controls actually worked over a review period, usually 3 to 12 months. A Type I report only confirms the controls existed on one particular day. For an AI agent vendor, insist on Type II. Anything less tells you nothing about how the company runs day-to-day. ISO/IEC 27001 ISO/IEC 27001 certifies that the vendor runs a formal information security management system (ISMS): documented risk assessments, defined controls, internal audits, and management review, all verified by an accredited certification body. It’s the most widely recognized security certification outside the US and often a hard procurement requirement in Europe, the UK, and the Gulf. A vendor with international customers should hold it alongside SOC 2, not instead of it. ISO/IEC 27701 (Privacy Information Management) ISO/IEC 27701 extends ISO 27001 with a privacy information management system (PIMS). It maps closely to GDPR concepts like controller and processor obligations, consent, and data subject rights. Almost every AI agent processes personal data at scale, and ISO 27701 is a decent signal that the vendor has built privacy into how it operates instead of delegating it to a policy PDF. ISO/IEC 42001 (AI Management Systems) ISO/IEC 42001 is the first certifiable international standard for AI governance. According to the International Organization for Standardization, it sets out requirements for building and maintaining an AI management system (AIMS): AI risk management, AI system impact assessments, lifecycle management, and oversight of third-party suppliers. For an AI agent vendor, this is the one that covers what SOC 2 and ISO 27001 don’t: how the vendor governs model behavior, training data, and the wider impact of autonomous systems. Worth Knowing: ISO 42001 certificates only started appearing in volume in 2024, and the accreditation ecosystem is still catching up. Check that the certificate came from a certification body accredited for ISO 42001 specifically (under ANAB or UKAS, for example), not just one accredited for ISO 27001. HIPAA (for Healthcare AI Agents) If the agent touches protected health information (PHI), the vendor has to comply with the HIPAA Privacy and Security Rules and sign a Business Associate Agreement (BAA). There’s no official HIPAA certification, so vendors prove compliance through third-party assessments, a SOC 2 with HIPAA mapping, or HITRUST CSF certification. A vendor that won’t sign a BAA has disqualified itself for healthcare work. PCI DSS (for Payment-Handling AI Agents) AI agents that process, store, or transmit cardholder data (think agents automating billing, refunds, or checkout) fall under PCI DSS. Ask for the vendor’s Attestation of Compliance (AOC) and check whether a Qualified Security Assessor validated it or the vendor assessed itself. The current version is PCI DSS 4.x, so an AOC that still references 3.2.1 is out of date. FedRAMP (for Government-Facing AI Agents) FedRAMP authorization is mandatory for cloud services sold to US federal agencies. Authorizations come at Low, Moderate, and High impact levels, and every authorized service appears on the public FedRAMP Marketplace. If a vendor claims FedRAMP status and isn’t in the Marketplace, either the claim is false or the service is still “in process,” and those are very different things. State and local buyers should look for StateRAMP instead. Worth Knowing: ISO 42001 Certificates ISO 42001 certificates only started appearing in volume in 2024, and the accreditation ecosystem is still catching up. Check that the certificate came from a certification body accredited for ISO 42001 specifically (under ANAB or UKAS, for example), not just one accredited for ISO 27001. HIPAA (for Healthcare AI Agents) If the agent touches protected health information (PHI), the vendor has to comply with the HIPAA Privacy and Security Rules and sign a Business Associate Agreement (BAA). There’s no official HIPAA certification, so vendors prove compliance through third-party assessments, a SOC 2 with HIPAA mapping, or HITRUST CSF certification. A vendor that won’t sign a BAA has disqualified itself for healthcare work. PCI DSS (for Payment-Handling AI Agents) AI agents that process, store, or transmit cardholder data (think agents automating billing, refunds, or checkout) fall under PCI DSS. Ask for the vendor’s Attestation of Compliance (AOC) and check whether a Qualified Security Assessor validated it or the vendor assessed itself. The current version is PCI DSS 4.x, so an AOC that still references 3.2.1 is out of date. FedRAMP (for Government-Facing AI Agents) FedRAMP authorization is mandatory for cloud services sold to US federal agencies. Authorizations come at Low, Moderate, and High impact levels, and every authorized service appears on the public FedRAMP Marketplace. If a vendor claims FedRAMP status and isn’t in the Marketplace, either the claim is false or the service is still “in process,” and those are very different things. State and local buyers should look for StateRAMP instead. Regulatory Frameworks AI Agent Vendors Must Comply With Certifications are voluntary. Regulations aren’t. A credible AI agent vendor should be able to explain, in writing, how it meets

One in five organizations has already suffered a breach traced back to shadow AI. Meanwhile, 63% of breached organizations either have no AI governance policy at all or are still drafting one. Below is a complete, copy-ready shadow AI policy template with twelve sections, plus guidance on adapting it for your company size, your industry, and the regulatory frameworks you answer to. The template assumes one hard truth up front: your employees are already using unapproved AI tools. A policy that pretends adoption hasn’t started yet fails on day one, so this one starts from the assumption that it has. What Is a Shadow AI Policy? A shadow AI policy is a formal document that defines how your organization discovers, evaluates, approves, and governs AI tools that employees adopt outside official IT channels. The term borrows from shadow IT, the older problem of unsanctioned software and hardware, but the AI version carries sharper risks: data pasted into a public model may be retained, used for training, or exposed in ways the organization can’t reverse. The policy does three jobs: it separates approved use from unapproved use, gives employees a fast and visible way to request new tools so the sanctioned route beats the workaround, and spells out what happens when someone crosses the line, including how the organization detects it and responds. Shadow AI Policy vs. General AI Acceptable Use Policy Many organizations already have an AI acceptable use policy (AUP) and assume it covers shadow AI. It usually doesn’t. An AUP tells employees how to behave inside approved tools. A shadow AI policy governs the tools themselves: which ones exist in your environment, which ones are allowed, and what happens with the rest. You need both. The AUP handles conduct; the shadow AI policy handles inventory and control. If you only have room for one document, fold the AUP’s data-handling rules into Section 6 of the template below. The Shadow AI Policy Template (Download Link and Copy-Ready Sections) We’ve created a compliance safe template for Shadow AI Policy, use the link below to create a copy and customize for your company: Download The Shadow AI Policy Template → Copy the sections below into your policy management system and replace the bracketed placeholders. The language is plain on purpose. Legalese gets skimmed. Section 1: Purpose and Scope This policy governs the acquisition, approval, and use of artificial intelligence tools, features, and services at [Company]. It applies to all employees, contractors, interns, and third parties with access to [Company] systems or data. It covers standalone AI applications, AI features embedded in existing software, browser extensions, AI agents, APIs, and personal AI accounts used for work purposes, on both corporate and personal devices. The purpose of this policy is to enable productive AI use while protecting [Company] data, customers, and legal obligations. This policy does not prohibit AI. It prohibits ungoverned AI. That last sentence matters. Employees read the purpose statement first, and it decides whether they see the policy as an enabler or a blocker. Section 2: Definitions and Terminology Shadow AI: any AI tool, feature, agent, or service used for work purposes without formal approval under this policy. Approved AI Tool: an AI tool listed in the Approved AI Tools Registry (Section 4) and used under a [Company]-managed account. Personal AI Account: an account on any AI service registered to a personal email address or paid for personally. AI Feature: AI functionality embedded within otherwise approved software (e.g., an AI assistant added to a project management tool), which requires separate evaluation. Sensitive Data: data classified as [Confidential] or [Restricted] under [Company]‘s data classification policy, including the prohibited data classes in Section 6. Define “AI feature” explicitly. Vendors now ship AI additions into already-approved SaaS products every month, and without this definition, those features inherit approval they never earned. Section 3: Roles and Responsibilities The CISO (or designated security lead) owns this policy, maintains the Approved AI Tools Registry, and runs the approval workflow. Department heads ensure their teams know the policy and surface tool requests rather than suppressing them. Legal and Compliance review tools that touch regulated data or fall under the EU AI Act, GDPR, HIPAA, or client contractual restrictions. IT operates detection and monitoring controls (Section 9). Every employee is responsible for using only approved tools for work, reporting unapproved AI use they discover, and requesting new tools through the workflow in Section 7 rather than adopting them directly. Insider Note: In organizations under roughly 200 people, the “CISO” in this section is often the same overworked IT lead who manages laptops. Name a real person, not a title that doesn’t exist yet. A policy that assigns duties to a phantom role is unenforceable, and auditors notice. Section 4: Approved AI Tools Registry [Company] maintains a registry of approved AI tools at [location/URL]. For each tool, the registry records: tool name and vendor, approved use cases, prohibited use cases, permitted data classes, account type (enterprise/team/individual), data retention and training settings, risk tier (Section 5), approval date, and next review date. Only tools listed in the registry may be used for work. Tools not listed are unapproved by default. The registry is reviewed [quarterly]. Keep the registry somewhere employees actually look, such as your intranet homepage or IT help center, not buried in a GRC platform they can’t access. An invisible registry recreates the problem the policy exists to fix. Section 5: Risk Tier Classification (Low, Medium, High) Each tool in the registry is assigned a risk tier. Low: the tool processes only public or internal non-sensitive data, runs under an enterprise agreement with training opt-out, and produces output that a human reviews before use. Approval by IT Security alone. Medium: the tool processes internal business data or connects to [Company] systems via API or integration. Approval by IT Security plus the data owner. High: the tool processes sensitive data, customer personal data, or regulated data; makes or influences consequential decisions (hiring, credit, medical, legal); or operates autonomously

Legacy threat modeling frameworks such as STRIDE were designed for software that behaves the same way over and over again. Agentic AI does no such thing. It can rewrite its own plan mid-task, call external tools, negotiate with other agents, and produce a different output from identical input. MAESTRO exists because none of the legacy threat modeling frameworks were built to handle that. MAESTRO stands for Multi-Agent Environment, Security, Threat, Risk, and Outcome. It is a seven-layer threat modeling framework created specifically for agentic AI systems, and it has become the closest thing the industry has to a standard method for reasoning about agent security. Understanding MAESTRO in the Context of Agentic AI What MAESTRO Stands For Each word in the acronym carries meaning. Multi-Agent Environment signals that the framework models entire ecosystems of interacting agents, not a single model behind an API. Security, Threat, Risk covers the core discipline: identifying attack surfaces, cataloging threats, and assessing likelihood and impact. Outcome is the part most frameworks skip. MAESTRO asks what an attack actually produces in the real world, because an autonomous agent with tool access turns a compromised prompt into a compromised action. The Origin of MAESTRO (Cloud Security Alliance) The Cloud Security Alliance published MAESTRO in February 2025. Its creator is Ken Huang, Co-Chair of the CSA AI Safety Working Groups and CEO of DistributedApps.ai. The CSA has since applied the framework publicly to real systems, including OpenAI’s Responses API and Google’s A2A protocol, which gives practitioners worked examples rather than just theory. The framework is openly published, and the CSA maintains an official companion tool, the MAESTRO Threat Analyzer, on GitHub. SOC 2, ISO 27001 and HIPAA done for you. Fixed fee, 100% audit pass rate. Audit-ready in 6 weeks. Not 6 months. Schedule Free Assessment Why Traditional Frameworks Fall Short for Agentic AI STRIDE, PASTA, LINDDUN, and OCTAVE all share a founding assumption: the system under analysis follows predictable logic with clearly defined boundaries. You draw the data flow diagram, mark the trust boundaries, and enumerate threats against components that behave deterministically. Agentic AI breaks every part of that assumption. Unique Security Challenges of Autonomous Agents Agents introduce three properties that legacy models cannot express. Non-determinism means the same input can produce different behavior, so you cannot enumerate execution paths in advance. Autonomy means the agent makes decisions and takes actions without a human approving each step, which collapses the usual assumption that a person sits between intent and execution. And in multi-agent systems there is often no stable trust boundary: agents delegate to other agents, consume tool outputs from external servers via protocols like the Model Context Protocol (MCP), and update their own memory and goals at runtime. The Gap Between Legacy Frameworks and Agent-Based Systems The practical consequence is coverage gaps. STRIDE has no category for goal manipulation, where an attacker gradually steers what an agent is trying to achieve. PASTA assumes attacker objectives and data flows are fixed, which fails for systems that learn and adapt during operation. LINDDUN addresses privacy but says nothing about agent collusion or memory poisoning. A threat model built purely on these frameworks will pass review and still miss the attacks that matter most in an agentic deployment. How MAESTRO Addresses Agentic-Specific Risks MAESTRO does not discard the older frameworks. It extends them with a layered reference architecture, an AI-specific threat catalog for each layer, and, critically, explicit analysis of how threats propagate between layers. That cross-layer lens is the framework’s real contribution, because most serious agentic incidents are chains: poisoned data influences a model, the model misleads an agent, and the agent takes an unauthorized action three layers away from where the attack started. The Seven Layers of the MAESTRO Framework MAESTRO decomposes any agentic system into seven layers, each with its own threat landscape. Layer 1: Foundation Models The core LLMs or other models the agents reason with. Threats here include adversarial examples, model extraction, backdoored weights, and jailbreaks that bypass safety training. If the model is a third-party API, supply chain risk lives at this layer too. Layer 2: Data Operations Everything the agent ingests, stores, and retrieves: training data, RAG pipelines, vector databases, and agent memory. Data poisoning and memory tampering are the signature threats at this layer, and they are especially dangerous because a poisoned memory persists across sessions and keeps shaping future decisions long after the initial attack. Layer 3: Agent Frameworks The orchestration software that turns a model into an agent: LangChain, CrewAI, AutoGen, custom planners, and tool-calling logic. Threats include prompt injection through tool outputs, insecure tool definitions, and manipulation of the planning loop itself. Layer 4: Deployment Infrastructure The servers, containers, and cloud services the agents run on. The CSA’s threat catalog here reads like traditional cloud security with an agentic twist: compromised container images carrying malicious agent code, Kubernetes orchestration attacks, denial of service against agent runtimes, and tampering with Infrastructure-as-Code templates that provision agent resources. Layer 5: Evaluation and Observability The systems that monitor, evaluate, and debug agent behavior. This layer is often forgotten, and attackers know it. The CSA specifically flags poisoning observability data: manipulating the telemetry fed to monitoring systems so that incidents stay hidden from security teams while malicious activity continues. Layer 6: Security and Compliance MAESTRO treats this as a vertical layer that cuts across all others: identity and access management, guardrails, policy enforcement, and compliance controls. Threats include permission escalation, guardrail bypass, and compromise of the security agents themselves in architectures where AI enforces policy on other AI. Layer 7: Agent Ecosystem The environment where agents interact with users, other agents, and marketplaces. This is where the genuinely novel threats live: agent impersonation, misleading agent capability cards, tool squatting, and collusion between agents to achieve outcomes no single agent was authorized to pursue. Insider Note: In real assessments, Layers 5 and 6 expose the maturity gap fastest. Most teams’ shipping agents can describe their model and their orchestration framework in detail, then