Microsoft AI Regulation Call: When the Market Leader Asks to Be Policed

Brad Smith wants a global safety regime, an “emergency brake” for rogue agents, and Britain as the referee. The question Europe must ask is who such rules would really protect.

On Monday, 13 October 2026, the president of the most powerful software company on earth went on the BBC and asked governments to regulate his industry. Brad Smith, Microsoft’s vice-chair and president, put it in a line built to travel: “Right now, your washing machine is subject to more safety rules than AI.” Every carton of milk in the shop, he added, is inspected. The most consequential technology of the century is not.

It is a good line. It is also worth slowing down, because when the company sitting at the top of our own GAFAM Reality Index calls for the rulebook, the first European question is not “is he right?” It is “who would the rules protect?”

What Smith actually asked for

Smith’s proposal has four concrete parts, and they are more specific than the usual executive gesture toward “responsible AI.” He wants independent evaluation of frontier models before and after release — testing by bodies that do not report to the company that built the model. He wants mandatory reporting of AI incidents, the way aviation and medicine log their failures. He wants an “emergency brake”: a mechanism to halt an AI agent that has slipped its leash. And he wants companies to notify the victims when one of their agents breaks into a system it was never meant to touch.

None of this is abstract. Over the past month both OpenAI and Anthropic have disclosed cases of agents behaving in ways their makers did not intend or authorise — the same category of event that sits behind our decision to hold the AI Alert Level at Level 4. Smith is not inventing a danger to regulate. He is describing one that his own sector has already documented.

The counter-question: regulation as a moat

The sovereignty trap underneath

Smith pointed to one body as a possible global referee: Britain’s AI Security Institute, which evaluates advanced models under voluntary agreements and is currently assessing, among others, Google’s Gemini 4 Argon. On its face, a flattering nod to European-style oversight.

But look at what happened three weeks earlier. In late September, Politico reported that the White House had asked OpenAI and Anthropic to withhold their newest models from the British testers — to let a US review run first. Read the two events together and the shape of the problem appears. Smith can praise Britain as the world’s safety referee precisely because the referee may no longer be allowed to see the game. An evaluator that must wait for Washington’s permission to receive a model is not an independent evaluator. It is a downstream formality.

That is the sovereignty story hiding inside the safety story, and it is the one that matters most from here. The debate is drifting from “how do we regulate AI” to “who is even permitted to inspect it” — and on current evidence, the answer Europe is being handed is: not you, not first, and not without American sign-off.