308. advisories under an OWASP LLM 2026 risk
ShieldMap · AI security, mapped
How the AI security standards connect.
- OWASP LLM risks (2026)
- 10
- Frameworks mapped
- 8
- Advisories under a risk
- 308
- Packages
- 77
- AI Act articles quoted
- 11
Pinch or Ctrl/⌘ + scroll to zoom, drag to move
The OWASP Top 10 for LLM Applications 2026, with OWASP's own mapping of each risk to 8 other frameworks, and the advisories on ShieldMap that fall under each risk through their CWE.
Check a lockfile against ShieldMap →Trace: which MCP servers install a vulnerable dependency →How these attacks work: 15 techniques →AI products and their makers →AI models, open and closed →Exploited in the wild →AI security incidents and exercises →Malicious packages, flagged extensions and skills →Supply-chain campaigns against AI packages →Malicious packages wearing AI names →Removed and flagged AI extensions →MITRE ATLAS matrix of attacks on AI →How the AI security standards connect →The EU AI Act's dates and articles, quoted →Malicious agent skills →Data as of 2026-11-22
| Risk | ASI | DSGAI | ATLAS | CWE | 600-1 | RMF | AICM | AIVSS | Advisories |
|---|---|---|---|---|---|---|---|---|---|
| LLM01:2026 Prompt Injection | ● | ● | ● | ● | ● | ○ | ● | ● | 0 |
| LLM02:2026 Sensitive Information Disclosure | ● | ● | ● | ● | ● | ○ | ● | — | 23 |
| LLM03:2026 Excessive Agency | ● | ● | ● | ● | ● | ○ | ● | ● | 8 |
| LLM04:2026 Supply Chain | ● | ● | ● | ● | ● | ● | ● | ○ | 5 |
| LLM05:2026 Data and Model Poisoning | ● | ● | ● | ● | ● | ○ | ● | ○ | 2 |
| LLM06:2026 Unbounded Consumption | ● | ● | ● | ● | ● | ○ | ● | ○ | 152 |
| LLM07:2026 Misinformation | ● | ● | ● | ● | ● | ● | ● | ● | 0 |
| LLM08:2026 Hidden Context Exposure | ● | ○ | ● | ● | ● | ○ | ● | ○ | 0 |
| LLM09:2026 Vector and Embedding Weaknesses | ● | ● | ● | ● | ● | ○ | ● | ○ | 0 |
| LLM10:2026 Improper Output Handling | ● | ● | ● | ● | ● | — | ● | ○ | 118 |
● at least one primary mapping · ○ supporting only · — none. Advisories are matched through a CWE OWASP lists as primary for that risk alone.
Frameworks and versions
- OWASP Top 10 for Agentic Applications (ASI) · 2026 (announced 2025-12-09)
- OWASP GenAI Data Security 2026 (DSGAI) · v1.0 (2026-03-17)
- MITRE ATLAS · content v2026.06 (format-version 6.0.0)
- MITRE CWE (Common Weakness Enumeration) · 4.20
- NIST AI 600-1 (Generative AI Profile) · v1.0 (July 2024)
- NIST AI RMF (AI 100-1) · v1.0 (2023)
- CSA AI Controls Matrix (AICM) · v1.1 (2026-06-22)
- OWASP AIVSS (AI Vulnerability Scoring System) · v0.8
Weaknesses behind the risks
- CWE-79 Cross-site Scripting (XSS)
- CWE-116 Improper Encoding or Escaping of Output
- CWE-200 Exposure of Sensitive Information to an Unauthorized Actor
- CWE-201 Insertion of Sensitive Information Into Sent Data
- CWE-212 Improper Removal of Sensitive Information Before Storage or Transfer
- CWE-285 Improper Authorization
- CWE-349 Acceptance of Extraneous Untrusted Data With Trusted Data
- CWE-359 Exposure of Private Personal Information
- CWE-400 Uncontrolled Resource Consumption
- CWE-494 Download of Code Without Integrity Check
- CWE-532 Insertion of Sensitive Information into Log File
- CWE-732 Incorrect Permission Assignment for Critical Resource
- CWE-770 Allocation of Resources Without Limits or Throttling
- CWE-798 Use of Hard-coded Credentials
- CWE-829 Inclusion of Functionality from Untrusted Control Sphere
- CWE-1426 Improper Validation of Generative AI Output
- CWE-1427 Improper Neutralization of Input Used for LLM Prompting
Regulation and other frameworks
- EU AI Act →8 application dates from Article 113 and 9 articles, quoted from the current consolidated text.
- NIST AI 600-1 and AI RMF →12 generative AI risks and 9 AI RMF categories, with the OWASP risks mapped to each.
- CoSAI Risk Map →36 AI security risks and 37 controls, with who is responsible for each.
Mappings from the OWASP Top 10 for LLM Applications 2026, Appendix A (OWASP GenAI Security Project), shared under CC BY-SA 4.0. Mappings restructured as data; the ATT&CK section and the AICM relevance text left out. ShieldMap links to these frameworks for navigation; it does not claim ShieldPi meets or certifies against any of them.