What is MITRE ATLAS? The knowledge base of attacks on AI systems, defined

MITRE ATLAS is a free, public knowledge base maintained by MITRE that catalogues adversary tactics, techniques, mitigations and real-world case studies for attacks on AI-enabled systems. It is laid out as a matrix in the style of MITRE ATT&CK, and security teams use it to threat-model, plan red-team work and classify AI incidents in shared terms.

How ATLAS is built

The ATLAS home page showed 16 tactics, 208 techniques, 40 mitigations and 73 case studies when this entry was written on 19 October 2026. MITRE says the content rests on observed real-world attacks and on realistic demonstrations by AI red teams and security groups. Its scope covers attacks on AI systems, abuse of AI capabilities and harmful autonomous behaviour enabled by AI.

The matrix can be filtered by platform (predictive AI, generative AI, agentic AI and enterprise IT) and by maturity, from feasible to demonstrated to realized. Each technique page carries an identifier, the tactics it serves, linked case studies and mitigations, and a versioned permalink. The LLM Jailbreak page, AML.T0054, for example, sits under Defense Evasion and Privilege Escalation, is rated Realized and dates from 25 October 2023.

How security teams use it

Government guidance points to it by name. The May 2026 joint guidance on agentic AI from Australian, US, Canadian, New Zealand and UK agencies recommends threat modelling against current taxonomies such as the OWASP GenAI Security Project and MITRE ATLAS. The same document cautions that both focus on LLM weaknesses, so some attack paths unique to agents may not yet be captured.

ATLAS also describes its own use in testing. Its AI Red Team mitigation, AML.M0035, tells teams to build threat vectors from ATLAS tactics, techniques and procedures and rank them by likelihood and impact. For a security leader, the practical value is a common vocabulary: an incident, a test finding and a control can all point to the same technique identifier.

Where it appears on cisonews.ai

The site's ATLAS technique index tags every entry in the AI incident log with the ATLAS technique or techniques it demonstrates, grouped by tactic. The tags are this site's own classification, applying ATLAS's published taxonomy to facts already reported on each incident page. The explainer how do you secure AI agents shows where ATLAS fits beside OWASP's lists.

Related terms

AI red teaming, prompt injection, jailbreak and data poisoning.