MITRE ATLAS

What ATLAS is

MITRE ATLAS stands for Adversarial Threat Landscape for Artificial-Intelligence Systems. It is a free, public knowledge base of adversary tactics and techniques aimed at AI systems, built by the same organisation behind MITRE ATT&CK and deliberately shaped like it. If you have used ATT&CK in a SOC, the layout will feel familiar: tactics are the attacker's goals, techniques are the ways they reach those goals, and each technique links to mitigations.

What makes ATLAS different is the subject. Its techniques cover things that do not exist in ordinary IT: tampering with training or retrieval data, crafting inputs that change a model's behaviour, extracting a model's behaviour through its API, compromising the AI supply chain and, more recently, attacks on agents that use tools. It also collects case studies of real incidents and research demonstrations, which show how attacks on AI systems unfold from start to finish.

Is it worth your time?

Yes, if you are learning AI security, application security for LLM features or detection engineering for AI systems. The OWASP Top 10 for LLM Applications tells a builder what to prevent. ATLAS helps you describe and plan: which techniques could affect this system, which mitigation covers each one, and how to write a finding in a shared vocabulary. The two work well together.

It is less useful as a first introduction to how language models work. Learn the basics of LLM apps, retrieval and tools first, then open ATLAS.

How to use it well

• Start with the matrix view and read the tactic names left to right to get the shape of an attack on an AI system.
• Read two or three case studies in full. They are the fastest way to understand why the techniques matter.
• When you threat model your own LLM app, pick the handful of techniques that genuinely apply and look up their mitigations, rather than trying to cover everything.
• Use technique IDs in your notes and reports so readers can look them up.

How much time does it take?

An afternoon is enough to learn the structure and read a few case studies. After that it is a reference you return to whenever you design, review or test an AI feature.

Free or paid?

Completely free. The data is also published openly on GitHub (the mitre-atlas/atlas-data repository) in machine-readable form, which is useful if you want to build tooling around it.

Common questions

Is ATLAS a checklist? No. Like ATT&CK, it is a map of known behaviour. Use it to decide what matters for your system, not to tick boxes.

Do I need to know ATT&CK first? It helps but is not required. The ideas of tactics, techniques and mitigations are explained on the site.

Is it only for red teams? No. Builders use it in threat models, defenders use it to plan monitoring, and managers use case studies to understand risk.