MITRE ATLAS Reference

101 techniques, 69 sub-techniques across 16 tactics

MITRE ATLAS is the adversarial-AI counterpart to ATT&CK: a public, evolving taxonomy of techniques used to attack AI systems, from poisoning training data to extracting models to bypassing prompt-injection defenses. Each tactic below groups techniques used to accomplish a specific adversarial goal against an AI system. Every page links to MITRE's authoritative source and includes practitioner context from GTK Cyber's AI security training.

Tactics

Reconnaissance 8 techniques

Resource Development 13 techniques

Initial Access 7 techniques

AI Model Access 4 techniques

Execution 6 techniques

Persistence 9 techniques

Privilege Escalation 4 techniques

Defense Evasion 15 techniques

Credential Access 6 techniques

Discovery 9 techniques

Lateral Movement 2 techniques

Collection 4 techniques

AI Attack Staging 6 techniques

Command and Control 3 techniques

Exfiltration 6 techniques

Impact 9 techniques

Looking for the broader ATT&CK framework?

See the MITRE ATT&CK reference for the full enterprise adversary taxonomy (697 techniques across 14 tactics).

AI red teaming, taught by people who do it.

GTK Cyber's hands-on AI security courses cover the ATLAS framework end-to-end.

View AI Security Courses