Microsoft CEO Nadella Calls for ‘Kill Switch’ on Advanced AI Models

Nadella's framing matters because it moves AI control from a research-safety debate into standard enterprise security practice: treat a model like an insider threat, assume compromise, and design containment in from the start. That lands directly on the firms deploying agents in production, not just the labs training them, and it implicitly sets an expectation that vendors ship pause and shutdown capability rather than bolt it on. The open question is whether a kill switch becomes a procurement requirement or stays a voluntary posture, and who is accountable when an agent is halted mid-task.