The governance of autonomous AI systems has been framed predominantly as a problem of model behavior: whether systems pursue intended objectives, avoid harmful outputs, and comply with stated policies. This paper argues that framing is insufficient. Governance is not a behavioral attribute that can be trained into a model or reconstructed through audit after the fact. It is an operational property of execution — one that must be preserved during runtime or it does not exist. The central claim is precise: when authority separation collapses during autonomous execution, no subsequent review, oversight mechanism, or audit trail restores institutional legitimacy. The violation occurred at the moment of execution. This paper develops that argument, specifies its invariant structural conditions, and draws out its consequences for institutions deploying autonomous systems at scale.
Ricardo Rubio Albacete (Sun,) studied this question.