Mapping AI-enabled cyber threats

This Anthropic Frontier Red Team report analyzes 832 banned accounts and 13,873 observations of AI-enabled misuse mapped to MITRE ATT&CK, introduces the ARiES risk score to quantify AI enablement, and finds a marked increase in medium/high-risk actors driven by agentic scaffolding that enables autonomous multi-stage operations (exemplified by the GTG-1002 cyber espionage campaign); it recommends extending ATT&CK to capture autonomous behaviors and adapting detections and safeguards accordingly.