← Home
Technical report · 2026
AI Runtime Threat Matrix
A threat taxonomy and control framework for autonomous AI agent systems, with the RAAI runtime control loop.
Track threats across the agent workflow.
The matrix organizes risks across identity, authorization, prompts and context, memory, tools and APIs, data flows, external systems, multi-agent interactions, autonomous execution, oversight, monitoring, and governance.
PurposeGive security teams a shared structure for identifying runtime attack paths and controls.
Control loopRAAI: Assess, Analyze, Authorize, Act, and Audit.
PublicationTechnical report, version 1.0, August 2026. DOI: 10.5281/zenodo.22063227.