With Great Capabilities Come Great Responsibilities: Introducing the Agentic Risk & Capability Framework for Governing Agentic AI Systems

Researchers introduce the Agentic Risk & Capability (ARC) Framework to help organizations govern autonomous AI systems safely. As agentic AI gains ability to execute code, access the internet, and modify files independently, new governance challenges emerge. The framework identifies three primary risk sources—components, design, and capabilities—and maps each to specific risks and technical controls.