Analysis
Microsoft president Brad Smith used a BBC podcast appearance to argue that the U.K., not the U.S. or the EU, is best positioned to write the rules for global AI safety, calling AI both a "weapon and a tool" and saying "the moment has arrived" for binding standards, TechCrunch reported Monday. His reasoning: Washington is too close to its own AI labs, and Brussels' regulatory instincts make U.S. companies defensive. "In Washington, D.C., people don't look at the U.K. and they don't look at London the way they look at Brussels," Smith said. "That's a strength."
Smith's timing matters, however: the Trump administration reportedly asked Anthropic and OpenAI in September to withhold new models from British testers, and Smith's own employer has a direct stake in whichever government ends up writing the rulebook its Azure AI business operates under. He called the White House's September agreement, in which AI companies made "morally binding" voluntary safety commitments, a "good step" that falls short, since voluntary measures only bind "the companies that are... responsible" -- his words, not a regulator's.
“Both ideas would require exactly the kind of binding enforcement Smith says voluntary commitments lack.”
What he's actually proposing is more specific than the usual safety rhetoric: a legal duty to disclose when a system is breached by a rogue AI agent, and outside evaluators approving products before release. Both ideas would require exactly the kind of binding enforcement Smith says voluntary commitments lack.
Room for disagreement: Smith, however, works for a company that just built one of the largest AI deployment footprints on the planet and has every incentive to pick the regulator it expects to be most workable. Britain's AI Security Institute, which he's implicitly nominating as the standard-bearer, has no enforcement power today -- it gets voluntary pre-release access to frontier models from U.S. labs, not legal authority over them. Dario Amodei and Sam Altman have both called for slowing AI development after incidents involving rogue agents, which gives Smith's framing political cover, but neither lab has committed to binding external approval of the kind he's floating.
The jurisdiction question isn't settled by picking the government with the lightest touch and the best manners. It's settled by which regulator can actually compel a frontier lab to change behavior, and right now that's none of them.