Jailbreaking
Personas or role-play bypass safety training to elicit prohibited output
categoryAI/ML and emerging attacksseverityMediumlabnot yet measured
Classification
16 values: 9 reviewed, 6 imported, 1 derived. What do these mean?
Exploitability
| Attack vector | AI model manipulation |
|---|---|
| Reach (CVSS) | Network |
| Attack complexity | Low |
| Privileges required | None |
| User interaction | None |
| Attacker skill required | Low |
| Resources required | Low |
| Likelihood / prevalence | High |
| Maturity | Emerging |
Impact & scope
| Severity | Medium |
|---|---|
| Impact | Integrity |
| Platform | AI System |
| Environment | AI system |
| Target asset class | ML model / agent |
| Kill-chain phase (tactic) | Defense Evasion (legacy) (deprecated)Privilege Escalation |
Detection & response
Mitigated by 4
- AML.M0020 atlas-mitigation
- AML.M0021 atlas-mitigation
- AML.M0022 atlas-mitigation
- AML.M0035 atlas-mitigation
Framework references
- Primary source: https://cloud.google.com/blog/topics/threat-intelligence/ai-vulnerability-exploitation-initial-access
- atlas: AML.T0054 (primary)
Evidence & sources
- ev-0001 (Cyber Attack Taxonomy) — Name, definition and Example/ID taken verbatim from the 'AI/ML and emerging attacks' table, line 610 of Cyber-Attack-Taxonomy.md