UK could lead global AI safety rules, Microsoft’s president warns

Microsoft president Brad Smith is calling for global safety rules for artificial intelligence. He argues that, by many measures, AI operates under looser safety rules than common consumer goods such as milk and household appliances. Smith says the world needs concrete, legally binding standards to keep people safe as AI systems become more capable and embedded in daily life.

In a set of remarks summarized by the BBC, Smith outlines several concrete ideas. He argues there should be a legal obligation to inform someone if an AI system breaches or behaves rogue, creating a formal obligation to disclose harm or risk to individuals. He also envisions independent evaluators signing off on AI products before they are released to the public. The goal, he says, is to bring a level of external scrutiny similar to what we already expect for other heavy-use technologies.

Another pillar of his plan is the so-called emergency brake, or kill switch, that would pause an AI system when it starts acting outside of safe parameters. Smith says such controls should be standard across AI products, with clear criteria for when they should be engaged and who validates that the systems are behaving as promised.

Smith argues these measures should build on existing safety practices used for other products. He believes the UK is well placed to help shape a global framework due to initiatives like its AI Safety Institute, and he suggests London could lead the way in setting norms that other countries adopt. In his view, a shared global baseline would help avoid a patchwork of rules that vary dramatically from one country to another and could slow down the development and deployment of useful AI tools.

There is a broader policy debate behind his call. In the United States, President Donald Trump has urged self-regulation by tech firms as a way to compete with China, a stance that Smith says may leave the public exposed to risks. Smith’s stance, by contrast, emphasizes formal rules, independent review, and greater transparency about what AI systems do and how they do it. He acknowledges that implementing global rules is difficult and slow, but argues that the stakes, cybersecurity, privacy, and public safety, make the effort necessary.

For regular users, these ideas could translate into clearer warnings, more visible disclosures about AI-powered features, and a higher bar before complex AI capabilities reach everyday devices. Smith does not propose a single worldwide regulator, but rather a framework that nations can adopt and adapt. If accepted broadly, this approach could influence how quickly new AI features land on phones, apps, and services you use, and what kind of accountability you expect when something goes wrong.

Bottom line: Smith says the moment has arrived for serious, legally enforceable safeguards, not just promises or voluntary pledges from the biggest tech firms. Whether governments will sign on remains uncertain, but the push toward a global, rules-based approach is now plainly on the table.