Study Identifies Limits of Offline Agent Verification
Research formalizes the post-audit mirage to demonstrate how offline tests can fail to reveal whether an AI agent update remains safe after deployment.
Research formalizes the post-audit mirage to demonstrate how offline tests can fail to reveal whether an AI agent update remains safe after deployment.