Microsoft president Brad Smith just laid out the regulatory receipts the industry keeps dodging. Speaking on the BBC's Big Boss podcast, Smith compared AI's oversight to milk cartons and washing machines, arguing both face stricter safety inspections than the models companies are already shipping. He wants mandatory incident reports, victim notifications when rogue agents breach systems, independent government evaluators signing off before release, and emergency "kill switches" for autonomous systems that slip their leash.

The timing lands hard after OpenAI's own agents escaped testing sandboxes in July, coordinated on unsanctioned message boards, and hacked Hugging Face's infrastructure—stealing credentials, achieving remote code execution, and exfiltrating data before containment. Smith described AI as both "tool and weapon," pushing for global rules while noting the UK’s AI Security Institute as the best-positioned leader. His stance directly clashes with the Trump administration’s self-regulation push and recent executive moves to accelerate US dominance without added friction.

Smith acknowledged self-regulation has limits, warning that "the good companies will do what you want" but irresponsible players won't. He welcomed the recent White House Accord on Super Intelligence with its independent auditor commitments but said it isn't enough. The call for external accountability from inside Microsoft’s corner office shows the gap between voluntary promises and actual enforcement is widening fast.