Malicious model files / pickle deserialization
Serialised model artifacts execute embedded payloads on load
categoryAI/ML and emerging attacksseverityVery Highlabnot yet measured
Classification
19 values: 11 reviewed, 5 imported, 3 derived. What do these mean?
Exploitability
| Attack vector | AI model manipulationSoftware vulnerabilitiesSupply chain |
|---|---|
| Reach (CVSS) | Local |
| Attack complexity | Low |
| Privileges required | None |
| User interaction | Active |
| Attacker skill required | Medium |
| Resources required | Low |
| Likelihood / prevalence | Medium |
| Maturity | Emerging |
Impact & scope
| Severity | Very High |
|---|---|
| Impact | AvailabilityConfidentialityIntegrity |
| Platform | AI System |
| Environment | AI system |
| Target asset class | ML model / agent |
| Kill-chain phase (tactic) | Execution |
Detection & response
Mitigated by 6
- AML.M0011 atlas-mitigation
- AML.M0013 atlas-mitigation
- AML.M0014 atlas-mitigation
- AML.M0016 atlas-mitigation
- AML.M0018 atlas-mitigation
- AML.M0023 atlas-mitigation
Framework references
- Primary source: https://hiddenlayer.com/innovation-hub/pickle-strike/
- atlas: AML.T0011.000 (primary)
Evidence & sources
- ev-0001 (Cyber Attack Taxonomy) — Name, definition and Example/ID taken verbatim from the 'AI/ML and emerging attacks' table, line 637 of Cyber-Attack-Taxonomy.md