MITRE ATLAS 2026 Update Expands Agentic AI Threat Taxonomy

MITRE released ATLAS v5.x updates through April 2026, expanding the framework to 16 tactics and 84 techniques with a major shift from model-centric attacks to execution-layer exposure specific to agentic AI systems. New case studies include SesameOp (OpenAI Assistants API exploitation).

ATLAS is the de facto threat modelling language for AI/ML adversarial tactics. The agentic focus reflects the industry-wide recognition that attack surfaces are shifting from model training to orchestration layers, tool-calling interfaces, and API chains.

Security teams should update threat models using ATLAS v5.x taxonomy. Review SesameOp and other case studies to understand emerging attack patterns on agentic orchestration layers.