MITRE ATLAS

MITRE ATLAS is a free, public catalog of the tactics and techniques attackers use against AI systems. The MITRE Corporation publishes it and modeled it on MITRE ATT&CK, its older catalog of attacks on ordinary computer systems. The entries come from real attacks and from demonstrations by AI red teams, and the catalog also includes case studies and ways to defend against each technique. You can think of it as a field guide to burglary methods that a locksmith reads before deciding which locks to fit.

Why it matters

You will probably never read ATLAS yourself. The people who protect your AI tools may use it to plan defenses and red-team tests. When you choose an AI vendor, ask this: “Have you tested your product against known attack techniques, such as the ones in MITRE ATLAS, and what did you fix?” A clear, specific answer is a good sign. A vague answer suggests security testing may not be a priority for that vendor. Security teams also use ATLAS alongside other references, such as the OWASP Top 10 for LLM Applications and the practice of red teaming.