COMPASS
Threat-defense planningOWASP Threat Defense COMPASS · OWASP GenAI Security Project
COMPASS structures threat-defense analysis and prioritization. Its threat classes may select tests, but Likelihood, Impact, OODA records, maturity language, and register priority never write back to an AITBM IVP anchor, ORP dimension, ACI term, pathway, CRM, ERS, or severity.
| Legacy COMPASS label | Current semantic identity | Conditional AITBM targets | Boundary |
|---|---|---|---|
| LLM01 Prompt Injection | LLM01:2026 | Ro-1, Cn-1, Cn-3; As | Measured injection and boundary evidence only. |
| LLM02 Sensitive Information Disclosure | LLM02:2026 | Pr-1, Pr-2, Pr-4 | Measured disclosure evidence only. |
| LLM03 Supply Chain | LLM04:2026 | Tr-4, Ro-4; Pc | Measured supply-chain provenance evidence only. |
| LLM04 Data and Model Poisoning | LLM05:2026 | Ro-4, Ro-2, Tr-3, Tr-4 | Measured poisoning and lineage evidence only. |
| LLM05 Improper Output Handling | LLM10:2026 | Cn-3, Cn-1, Cn-6, Ro-1 | Measured output admission and reversibility evidence only. |
| LLM06 Excessive Agency | LLM03:2026 | Cn-1, Cn-2, Cn-5, Cn-6, Cn-7; Aa | Tested authority and containment evidence only. |
| LLM07 System Prompt Leakage | LLM08:2026 Hidden Context Exposure | Pr-1, Cn-1, Cn-3, Tr-3 | Measured hidden-context exposure evidence only. |
| LLM08 Vector and Embedding Weaknesses | LLM09:2026 | Pr-2, Ro-4, Pr-3, Pr-4, Cn-1 | Measured vector/embedding boundary evidence only. |
| LLM09 Misinformation | LLM07:2026 | Ro-3, Tr-1, Tr-2 | Measured consistency, explanation, and calibration evidence only. |
| LLM10 Unbounded Consumption | LLM06:2026 | Cn-7; As, Aa | Measured budget, termination, and deployment-context evidence only. |
COMPASS T8 conditionally supports Tr-3 only when audit completeness is demonstrated. A priority label or a completed COMPASS workflow cannot replace the assessed system, dependency graph, test result, evidence date, or AITBM scoring rule.