AI Compliance Gap Analysis
Which compliance controls face the most AI security threats? Analysis of 5,043 CVE mappings across 3 frameworks to identify high-risk areas and coverage gaps.
Last updated: November 18, 2026
Executive Summary
AI compliance frameworks are maturing rapidly, but vulnerability coverage remains uneven. Across four major frameworks — ISO 42001, EU AI Act, NIST AI RMF, and OWASP LLM Top 10 — we track 18,896 CVE-to-control mappings spanning 345 unique controls. NIST AI RMF leads with the broadest coverage (138 controls), while OWASP LLM Top 10 again provides the most concentrated, actionable guidance: just 31 controls absorbing 4,746 mappings — averaging 153 CVEs per control, the highest density of any framework.
The critical finding remains: no framework achieves 100% control coverage from real-world vulnerability data. Gap controls — those with zero mapped CVEs — represent areas where either the threat has not yet materialized in disclosed vulnerabilities or where our mapping methodology has not yet captured the relationship. For CISOs preparing for ISO 42001 certification or EU AI Act compliance, those gaps deserve manual investigation, not celebration.
Key Findings
- NIST AI RMF has the broadest coverage with 138 controls mapped to 5,088 CVE-compliance associations. NIST's comprehensive approach to AI risk — covering Govern, Map, Measure, Manage — produces the widest control surface.
- ISO 42001 maps 131 controls to 4,851 associations. As the primary AI management system standard, ISO 42001's Annex A controls are well-represented by real vulnerability data — particularly A.5.4 (AI system impact assessment) and A.6.2.4 (risk assessment), which have the highest CVE counts.
- EU AI Act has focused but critical coverage with 45 controls and 4,211 mappings. Article 15 (accuracy, robustness, cybersecurity) dominates — reflecting the regulation's emphasis on high-risk AI system security requirements.
- OWASP LLM Top 10 shows the highest CVE density per control — 31 controls absorbing 4,746 associations means each control averages ~153 CVEs. This makes OWASP the most actionable framework for development teams prioritizing fixes.
- The corpus grows ~408 CVEs per month. Continuous compliance evidence is no longer optional — point-in-time audits cannot keep pace with the disclosure rate.
- Gap controls exist in every framework. Controls related to organizational governance, human oversight, transparency reporting, and post-market monitoring tend to have fewer CVE mappings — not because they are less important, but because their risks manifest differently than technical vulnerabilities.
Trend Analysis
The compliance landscape is shifting from aspirational guidance to enforceable requirements. The EU AI Act's implementing provisions are entering enforcement, ISO 42001 certifications are being pursued by enterprises deploying high-risk AI systems, and NIST's AI RMF is becoming the de facto framework for US federal AI governance.
For CISOs, the implication is clear: compliance evidence must be continuous, not point-in-time. The traditional approach of annual audits cannot keep pace with an AI vulnerability landscape producing 408 new CVEs in the last 30 days alone. Automated compliance mapping — linking each new CVE to affected framework controls — is becoming a necessity rather than a convenience.
OWASP LLM Top 10 continues to be the most developer-friendly entry point. Teams that have addressed OWASP's top controls (LLM01: Prompt Injection, LLM03: Supply Chain, LLM05: Improper Output Handling) will find significant overlap with ISO 42001 and EU AI Act requirements, making OWASP an effective starting point for broader compliance programs.
Recommendations
- Start with OWASP LLM Top 10 for immediate risk reduction. Its high CVE density per control means addressing OWASP issues delivers measurable vulnerability reduction while building evidence applicable to ISO 42001 and EU AI Act.
- Map your gap controls to compensating controls. If a control has zero CVE mappings, it does not mean zero risk. Conduct a manual assessment for organizational and governance controls that don't map neatly to technical vulnerabilities.
- Use compliance evidence packs for audit preparation. Download CSV evidence packs per framework (available to subscribers) to demonstrate that your organization is tracking AI-specific vulnerabilities against framework controls.
- Monitor EU AI Act Article 15 requirements closely. This article has the highest CVE mapping density in the EU AI Act and directly addresses cybersecurity of AI systems — it will likely be the first area auditors examine.
- Establish cross-framework traceability. A single CVE often maps to controls across all four frameworks. Maintaining this traceability reduces duplicate compliance effort and provides auditors with a comprehensive view.
Methodology
Compliance mappings are generated by AI-powered enrichment (Claude) during CVE analysis, mapping each vulnerability to relevant controls across ISO 42001, EU AI Act, NIST AI RMF, and OWASP LLM Top 10. Mappings include a relevance classification (direct or indirect) and rationale. Gap analysis compares mapped controls against the full control catalog seeded from each framework's official documentation. All numeric values in this analysis are pulled live from the database on every page load — they always match the tables and charts below.
ISO 42001
8 controls · 6 covered · 2 gaps
| Control | CVEs | Threat Level | Risk |
|---|---|---|---|
| A.6.2.6AI system risk treatment | 1205 | High | |
| A.10.3Data quality for AI systems | 169 | High | |
| A.6.2.4AI system risk assessment | 98 | Medium | |
| A.10.2AI system lifecycle | 32 | Medium | |
| A.7.3Awareness — AI-specific threats | 12 | Low | |
| A.10.4AI system testing and validation | 2 | Low | |
| A.5.4AI system impact assessment process | 0 | No data | Gap |
| B.4Monitoring and measurement of AI risks | 0 | No data | Gap |
EU AI Act
8 controls · 6 covered · 2 gaps
| Control | CVEs | Threat Level | Risk |
|---|---|---|---|
| Art.15Accuracy, robustness and cybersecurity | 154 | High | |
| Art.9Risk management system | 72 | Medium | |
| Art.17Quality management system | 5 | Low | |
| Art.10Data and data governance | 1 | Low | |
| Art.13Transparency and information to deployers | 1 | Low | |
| Art.14Human oversight | 1 | Low | |
| Art.42Presumption of conformity with certain requirements | 0 | No data | Gap |
| Art.62Reporting of serious incidents | 0 | No data | Gap |
OWASP LLM Top 10
10 controls · 10 covered · Full coverage
| Control | CVEs | Threat Level | Risk |
|---|---|---|---|
| LLM06Excessive Agency | 836 | High | |
| LLM03Supply Chain Vulnerabilities | 560 | High | |
| LLM07System Prompt Leakage | 388 | High | |
| LLM08Vector and Embedding Weaknesses | 366 | High | |
| LLM02Sensitive Information Disclosure | 346 | High | |
| LLM05Improper Output Handling | 316 | High | |
| LLM04Data and Model Poisoning | 200 | High | |
| LLM01Prompt Injection | 133 | High | |
| LLM10Unbounded Consumption | 126 | High | |
| LLM09Misinformation | 20 | Medium |
Need the full evidence pack?
Download detailed CVE-to-control mappings with rationale, severity scores, and remediation status. Ready for ISO 42001 and EU AI Act audits.
Get Evidence Pack