MITRE ATLAS v2: the AI attack taxonomy updated with real case studies
In one sentence MITRE releases ATLAS v2 (Adversarial Threat Landscape for AI Systems), an expanded taxonomy of AI system attack techniques with real adversarial ML case studies and mapping to MITRE ATT&CK.
MITRE ATT&CK is the standard catalog security teams use to describe how attackers target IT systems. MITRE ATLAS does the same for artificial intelligence systems: it lists and classifies all known techniques for attacking, deceiving, or manipulating AI models.
Version 2 adds documented real case studies: adversarial attacks on production image recognition systems, evasion of ML-based malware detectors, manipulation of recommendation systems via data poisoning.
For a security team, having a shared language to describe "we experienced a reconnaissance attack on an ML model" provides the same advantage ATT&CK gave for traditional threats: it facilitates communication, threat modeling, and incident response.
Companies
MITRE
Tools
MITRE ATLAS
Tags
Sources