A Five-Plane Reference Architecture for Runtime Governance of Production AI Agents

In three linesarXiv paper presenting a five-plane reference architecture for runtime governance of production AI agents. Decomposes into reasoning plane and four enforcement planes (network, identity, endpoint, data). Addresses risks from delegation chains and composite actions, with implementation and evaluation against 7 agent threats across 5 workflows.

Summary generated by Claude — human-verified