Overview
MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a knowledge base of real-world adversary tactics and techniques targeting AI systems — modelled on the format of MITRE ATT&CK, but for machine learning.
How it is organized
Like ATT&CK, ATLAS is a matrix of tactics (the adversary's goal at each stage) and techniques (how they achieve it), spanning the ML lifecycle:
- Reconnaissance and resource development against a target model
- ML attack staging — crafting adversarial inputs, poisoning data
- Model access, exfiltration, and evasion
- Impact on the AI-enabled system
Each technique includes real case studies of attacks observed in the wild.
Why it matters
ATLAS gives defenders a shared vocabulary for AI threats and a way to reason about coverage — "which ATLAS techniques can our system be hit by, and what detects or mitigates them?" It pairs naturally with the OWASP LLM Top 10 (risks) and NIST AI RMF (governance).
References
See the primary source below.