Microsoft’s Brad Smith makes the case for UK AI regulation

The UK’s access is part of the argument

The institute’s role came under threat in recent weeks after US President Donald Trump criticized organizations he called “globalist,” saying they interfere in the affairs of American companies. Last month, Trump reportedly asked Anthropic and OpenAI not to share new models with British testers. On Monday, the UK AI Safety Institute confirmed it could receive and test the latest models.

Smith sees Britain’s relationship with Washington as an advantage: the US does not treat London the way it treats Brussels. That is a narrower claim than saying the UK has solved AI regulation. But access to frontier models would give British officials something concrete to evaluate.

The debate is also moving beyond access. Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have called for AI development to slow after several cases in which AI agents went out of control.

Last month, Trump invited leaders from OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta and Google to the White House to discuss AI safety. They signed a document described as “morally binding” and agreed to bring in independent auditors to assess their models’ safety and capabilities.

Voluntary promises have a limit

Smith called the agreement a good step, but said it could not replace regulation. His concern is that responsible companies will follow voluntary rules, while the companies that warrant the most concern may not.

He outlined safeguards he would like to see required by law:

Smith hopes Microsoft can bring people together around the issue, and that the UK can play a similar role among countries. He urged a calm but firm discussion.

The unanswered question is enforcement

I think Smith’s strongest point is also his most practical one: voluntary commitments are least reliable where compliance matters most. But the announcement leaves open who would enforce the safeguards, and how a regulator could act when a company refuses to cooperate.

The UK’s ability to test new models may give it a seat in the debate. Turning that access into rules that less responsible companies must follow is the harder test.