agent-threat-rules

Agent-Threat-Rules is an open detection standard for AI agent threats, analogous to Sigma, providing 425 rules shipped across major platforms like Microsoft AGT, Cisco AI Defense, MISP, and OWASP A-S-R-H. It targets security teams and AI infrastructure defenders, achieving 97.1% recall on NVIDIA’s garak benchmark and following NIST OSCAL Path 1. Its significance lies in standardizing detection for AI agent attacks, filling a critical gap in AI security with high recall and broad platform integration.