Microsoft's Brad Smith Says the UK Is Best Placed to Lead AI Regulation

Microsoft President Brad Smith says the UK is better positioned than Washington or Brussels to lead global AI regulation, pointing to the UK's AI Security Institute as evidence it can do the job. Smith made the case in a BBC podcast interview published October 12, 2026.

Smith Calls AI 'A Weapon and a Tool'

Smith described artificial intelligence as "a weapon and a tool" that puts global cybersecurity at risk, and said the moment has arrived to put safety rules in place internationally rather than country by country. He proposed three specific obligations for AI companies: a legal requirement to disclose when a system has been breached by a rogue AI agent, a requirement that outside evaluators approve a model before it ships, and support for kill-switch mechanisms that let humans pause or slow an AI system if control is lost. He argued that consumer products as mundane as milk and washing machines currently face more regulatory oversight than AI companies do.

Smith's Case for UK-Led AI Regulation Over Washington and Brussels

Smith's argument for UK leadership rests on geopolitical positioning more than technical capacity. He said Washington does not view London through the same skeptical lens it applies to Brussels on regulatory matters, which he called an advantage the UK can use to broker agreement between the US, the EU, and AI companies. He closed by calling for the parties to work toward bringing people together on shared global rules.

Britain's AI Security Institute Already Tests Frontier Models Before Release

The UK's case for a leadership role isn't just rhetorical. Its AI Security Institute, known as AISI, launched in 2023 and already receives pre-deployment access to frontier models from major US labs, giving it a working testing relationship that other regulators don't have. An unnamed UK minister told Reuters on October 12 that AISI can access and test the latest models, a confirmation that came the same day Smith's remarks were published and amid reporting that access has recently been contested.

The White House Told US Labs to Hold Models Back From UK Testers

That contest became public on September 24, 2026, when Politico reported that the White House's Office of the National Cyber Director asked OpenAI and Anthropic to withhold their newest models from AISI until the US government had reviewed them first. A senior administration official told Politico the policy applies to every new frontier model from American companies, on the grounds that US systems should be reviewed before international partners get access. A British official confirmed the substance of the report the next day. Anthropic has already complied, Politico reported, withholding a model described as Mythos 5.1 from the UK institute. President Trump has separately warned against what he calls "globalist" organizations interfering with American AI companies.

Smith Says Last Month's White House Safety Pledge Falls Short

Smith's push for binding rules follows a White House summit last month where Trump hosted executives from OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta, and Google to discuss AI safety. The group signed a document described as "morally binding," committing to work with independent auditors to assess their models' safety and capabilities. Smith called that document a good step but said it stops well short of the legal obligations he is proposing. Anthropic's Dario Amodei and OpenAI's Sam Altman have each separately called for slowing AI development following incidents involving rogue AI agents, a position that lends weight to Smith's argument that voluntary commitments from responsible labs won't constrain the ones that aren't.

Why This Matters for Engineering Leadership

Engineering leaders who build on frontier models should treat the AISI access dispute as an early signal of how model access can fragment by jurisdiction. If the US government keeps gating which countries' safety institutes get pre-deployment access to new models, the models available to a UK-based or UK-regulated team could lag what a US team gets, with no public notice of the gap beyond press reporting. Teams planning model upgrades or compliance timelines around specific frontier releases should build in a buffer for jurisdiction-specific availability instead of assuming simultaneous global rollout. Smith's proposed rules, breach disclosure obligations, mandatory outside evaluation before release, and kill-switch support, also preview the kind of contractual and compliance terms that enterprise AI vendors may eventually be required to carry, regardless of whether any single government adopts them first. Leaders negotiating multi-year AI vendor agreements now have a concrete reason to ask what happens to their access and support if regulatory requirements diverge between the US, UK, and EU before the contract ends.

FAQ