Proof of Execution: Runtime Verification for Governed AI Agent Actions

James Rhodes, George Kang

Agent systems increasingly execute rather than advise. When an AI agent queries regulated data, invokes effectful tools, and mutates persistent state, correctness is not captured b…