AML.CS0019
PoisonGPT
- Type
- Exercise
- When
- 2023-07
- Actor, as ATLAS names it
- Not stated
- Target
- HuggingFace Users
- ATLAS attributes it to Mithril Security (security evaluation).
Techniques, in order
- 1Acquire Public AI Artifacts: ModelsResource Development · AML.T0002.001
- 2Manipulate AI Model: Poison AI ModelAI Attack Adaptation · AML.T0018.000
- 3Verify AttackAI Attack Adaptation · AML.T0042
- 4Publish Poisoned AI Artifacts: ModelsResource Development · AML.T0115.001
- 5Supply Chain Compromise: Model RepositoryInitial Access · AML.T0010.003
- 6Erode AI Model IntegrityImpact · AML.T0031
- 7External Harms: Reputational HarmImpact · AML.T0048.001
How these attacks work
From MITRE ATLAS 2026.10 (Apache-2.0). ShieldMap lists who and what was involved and the techniques used, and links to ATLAS for the write-up; it does not copy the steps. Attribution is ATLAS's, not ShieldPi's.