Autonomous Action Runtime Management (AARM) Working Group

The Autonomous Action Runtime Management (AARM) Working Group is developing an open, vendor-neutral specification for securing AI-agent actions before they reach connected systems. AARM addresses risks traditional security tools were not designed to manage, including irreversible actions, high-speed execution, excessive privileges, prompt injection, intent drift, and threats that emerge across a sequence of otherwise permitted actions. The specification defines how runtime security systems should intercept actions, evaluate them against policy and user intent, enforce appropriate controls, and create tamper-evident records for accountability.

Members help shape AARM’s threat model, conformance requirements, implementation guidance, architecture patterns, benchmarks, and public review process. The group welcomes security practitioners, researchers, technology providers, buyers, and end users who want to establish a shared foundation for building, evaluating, and advancing runtime security for AI agents.