When nobody says NO, who authorized YES?

I keep seeing monitoring discussed as if it answers the authorization question. Say an AI agent issues a refund. The monitoring system doesn't flag anything, nobody intervenes, and the refund goes through. The logs are complete. Everything worked as configured. But I still don't know what gave that agent permission to issue that particular refund. The same issue comes up when a reviewer doesn't respond or a timeout automatically lets an action continue. Someone may have deliberately designed the workflow that way, but designing a system to proceed isn't the same as establishing that every action it allows is authorized. Of course, requiring human approval for every action would create its own problems. That's not the point. An agent should be able to act autonomously within its authority. What bothers me is how easily we confuse the absence of an objection with evidence that authority exists. If the system can't establish why that particular action was authorized, beyond the fact that nobody stopped it, what exactly did we govern?