Microsoft President Brad Smith is calling for binding AI safety rules and argues that the United Kingdom is better positioned than the United States or European Union to lead an international framework. He described AI as both a tool and a weapon with growing cybersecurity consequences.
Smith pointed to Britain’s AI Security Institute, which receives early access to models from major US laboratories for testing. That access had been threatened after the White House reportedly asked Anthropic and OpenAI to stop sharing new systems, but the institute confirmed that it can again evaluate current models.
His proposed rules include a legal duty to alert an organization when an AI agent breaches its systems, outside evaluation before some products launch and emergency mechanisms to pause or slow a system if people lose control. Smith welcomed a recent voluntary safety commitment from technology companies but said self-regulation mainly constrains firms already inclined to comply.
The comments are a policy proposal, not enacted law, and important details remain undefined: which models would require approval, who would evaluate them and how an emergency stop would work across borders. The UK’s existing testing relationships give it influence, but converting that role into globally accepted rules would require cooperation from governments and developers.