New Framework Verifies Governed AI Agent Actions at Runtime

As AI agent systems move beyond mere advice to actively executing tasks, especially those involving sensitive data or state changes, ensuring their actions are authorized and verifiable becomes critical. Traditional methods of checking plausible outputs are insufficient when agents interact with regulated data, invoke tools, or modify persistent systems. The "Proof of Execution" (PoE) framework addresses this by formalizing runtime verification. It defines an execution as a contract, an event stream, and a replay context, establishing five validator-checkable invariants and semantic guarantees covering authorization, compliance, integrity, and replayability. PoE's soundness is cryptographically proven, meaning any adversary violating its guarantees would require signature forgery or hash collision. A prototype shows minimal overhead, adding only milliseconds to execution and compressing traces efficiently. While not replacing consensus or TEEs, PoE provides a crucial layer for attesting governed AI actions under contract.