CVSS
Vulnerability scoring (prior art)Common Vulnerability Scoring System (CVSS) · FIRST.org (CVSS Special Interest Group)
CVSS is the established 0-10 severity standard for discrete software vulnerabilities. AITBM is a complementary AI-system assessment framework, not a successor to CVSS; it adds fairness, transparency, AI-privacy, poisoning, drift, agent-identity, deployment-context, and evidence-confidence dimensions for risks not represented by a CVSS Base score.
| CVSS Metric Group | Primary AITBM Sub-Metrics | Evidence Use / Notes |
|---|---|---|
| Vulnerable System Confidentiality (VC/C) | Pr-1Pr-2Pr-4Cn-3 | Loose; CVSS has no membership-inference / extraction concept |
| Vulnerable System Integrity (VI/I) | Ro-3Ro-4Cn-1 | Loose; no probabilistic / poisoning corruption in CVSS |
| Subsequent System / Scope (SC-SI-SA / S) | Cn-5 | Partial; ORP Cp models multi-agent blast radius, not a binary flag |
| Attack Vector / Complexity / Requirements (AV/AC/AT) | Ro-1 | Partial; AI exploitability is empirical (attack-success-rate) |
| Exploit Maturity (E, Threat group) | ACI Tf | Inverted; CVSS ages the exploit, ACI ages the defender's evidence |
| Environmental group (Security Reqs, Modified Base) | ORP CRM + architecture-specific IVP weights | Closest analogue; applies deployment-specific modifications |
| Supplemental: Safety / Automatable / Recovery (v4.0) | ORP Aa, Cp, Rf | Gestural and non-scoring in CVSS; first-class scoring inputs in AITBM |
| (No CVSS metric) | Fa-1Fa-4Tr-1Tr-4 | No correspondence; CVSS has no bias or explainability axis |
| (No CVSS metric) | Ro-2 | No correspondence; CVSS cannot represent drift or assessment provenance |
Key findings
- Scope distinction: a CVSS Base score describes intrinsic vulnerability severity and is stable unless vulnerability facts change; CVSS v4.0 Threat and Environmental metrics can reflect exploitation state and deployment context. CVSS does not provide AITBM's AI-system dimensions, and its standardized formula is not architecture-weighted for AI systems.
- A CVSS Base score is intentionally stable, while Threat and Environmental metrics may change with exploitation and deployment context. AITBM's ACI answers a different question by decaying confidence when evidence supporting an AI-system assessment becomes stale.
- Complementary, not competitive: CVSS remains correct for conventional CVEs inside an AI stack (an unpatched serving-stack CVE even feeds AITBM's ORP As); AITBM scores the AI-specific risk layer that has no CVE, patch, or static severity. Never average a CVSS Base score with an ERS.
- Worked contrast (an EchoLeak-class scenario): Microsoft assigned CVSS 9.3 while the NVD Base score is 7.5. Those scores describe the vulnerability under their stated vectors; Threat and Environmental values can vary. The illustrative AITBM scenario yields ERS 7.1, with an evidence-age interval of 6.7->7.6, and models remediation to 4.0.