Backdoor / sleeper-agent poisoning
Trigger patterns implanted so the model misbehaves only on attacker inputs
categoryAI/ML and emerging attacksseverityHighlabnot yet measured
Example: Sleeper Agents (2024)
Classification
16 values: 9 reviewed, 6 imported, 1 derived. What do these mean?
Exploitability
| Attack vector | AI model manipulation |
|---|---|
| Reach (CVSS) | Local |
| Attack complexity | High |
| Privileges required | High |
| User interaction | None |
| Attacker skill required | High |
| Resources required | High |
| Likelihood / prevalence | Low |
| Maturity | Emerging |
Impact & scope
| Severity | High |
|---|---|
| Impact | Integrity |
| Platform | AI System |
| Environment | AI system |
| Target asset class | ML model / agent |
| Kill-chain phase (tactic) | AI Attack StagingPersistence |
Detection & response
Mitigated by 4
- AML.M0005 atlas-mitigation
- AML.M0008 atlas-mitigation
- AML.M0013 atlas-mitigation
- AML.M0035 atlas-mitigation
Framework references
- Primary source: https://arxiv.org/abs/2401.05566
- atlas: AML.T0018 (primary)
Evidence & sources
- ev-0001 (Cyber Attack Taxonomy) — Name, definition and Example/ID taken verbatim from the 'AI/ML and emerging attacks' table, line 616 of Cyber-Attack-Taxonomy.md