MITRE ATLAS is a public knowledge base of tactics and techniques used to attack AI-enabled systems. Its name stands for Adversarial Threat Landscape for Artificial-Intelligence Systems. It is modelled on MITRE ATT&CK, which security teams already use for conventional threats.
Why It Matters
AI systems fail in ways that ordinary threat models miss: poisoned training data, adversarial inputs, model theft and prompt-based attacks. ATLAS gives teams a shared vocabulary for these attacks, so that red teams, defenders and managers describe the same thing in the same way.
How It Is Organised
Like ATT&CK, ATLAS organises attacks into tactics, which are the attacker's goals, and techniques, which are the ways they achieve them. It also collects case studies of real incidents and demonstrations.
How to Use It
- Threat modelling: walk through the tactics and ask which apply to your AI system
- Red team planning: choose test cases that map to techniques
- Reporting: tag each finding with a technique so readers can look it up
- Defence planning: check which mitigations cover which techniques
ATLAS and the OWASP LLM Top 10
They complement each other. The OWASP Top 10 for LLM Applications is a ranked list of the main risks for LLM apps. ATLAS is a broader catalogue of adversary behaviour across AI systems. Many teams plan with ATLAS and test against the OWASP list.
Our AI Security course teaches ATLAS in the red-teaming methodology module, and our LLM pentesting and AI red teaming services map findings to it.
