MITRE ATLASmitigation
AML.M0008 Validate AI Model
- Created
- 2023-04-12
Description
Validate that AI models perform as intended by testing for backdoor triggers, potential for data leakage, or adversarial influence.
Monitor AI model for concept drift and training data drift, which may indicate data tampering and poisoning.
Related
Mitigates · MITRE ATLAS (8)
- AML.T0057 LLM Data Leakage technique
- AML.T0010.003 Model sub-technique
- AML.T0018 Manipulate AI Model technique
- AML.T0018.000 Poison AI Model sub-technique
- AML.T0018.001 Modify AI Model Architecture sub-technique
- AML.T0020 Poison Training Data technique
- AML.T0043 Craft Adversarial Data technique
- AML.T0043.004 Insert Backdoor Trigger sub-technique