AI Models & Assistants / Knowledge library
AI Threat Modeling
Maps assets, trust boundaries, attackers, misuse cases, and mitigations for an AI system
Why learn AI Threat Modeling?
It prioritizes red-team tests and controls according to credible attackers and system impact.
The problem it solves
It prevents security work from becoming an unbounded list of attacks with no deployment context.
Where it came from
AI threat modeling extends software security analysis to model inputs, outputs, training data, tools, retrieval sources, and human workflows. The relevant threats depend on the deployed system, not only the foundation model.
Build your understanding
Work through these exercises in order. Try the task before opening a reference.
- Define AI Threat Modeling and distinguish it from a nearby concept.
- Choose a realistic task where AI Threat Modeling is appropriate.
- Identify one failure mode or misuse of AI Threat Modeling.
- Interpret a small result produced with AI Threat Modeling.
- Explain one limitation or tradeoff of AI Threat Modeling.
Documentation and sources
- MITRE ATLAS (atlas.mitre.org)