Microsoft’s Brad Smith Backs UK Led Global AI Safety Rules
Microsoft's Brad Smith says the UK should lead binding global AI rules, with outside audits, breach alerts and kill switches replacing self-regulation.
Summary
Microsoft President Brad Smith says AI companies face less oversight than milk or washing machines, despite AI functioning as both a weapon and a tool that threatens global cybersecurity. He wants urgent, binding global safety rules and says the UK can lead because Washington views London differently from Brussels. Britain’s AI Security Institute, launched in 2023, receives major models from US frontier labs for testing before deployment. President Donald Trump warned that globalist organizations interfere with American companies and reportedly asked Anthropic and OpenAI in September 2026 to withhold new models, but AISI confirmed on October 12 that it can access and test the latest systems.
Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman have urged slower AI development after several incidents involving rogue agents. In September, Trump hosted OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta and Google at the White House, where they signed a morally binding commitment to use independent auditors for model safety and capability assessments. Smith calls that progress but insufficient because voluntary rules may bind responsible companies while leaving others unchecked. He supports mandatory notifications when rogue AI breaches an organization, outside approval before product releases and kill switches that can pause or slow systems if humans lose control, with Microsoft and the UK helping coordinate international action.
Positives
- Britain’s AI Security Institute has tested major US frontier models before deployment since its 2023 launch.
- AISI confirmed on October 12 that it retains access to the latest models despite reported US pressure.
- OpenAI, Anthropic, Nvidia, AMD, SpaceX, Meta and Google committed to independent model assessments at the White House.
- Smith supports mandatory breach notifications, external predeployment approval and kill switches for systems that escape human control.
Risks & concerns
- AI companies receive less oversight than milk and washing machines despite cybersecurity risks, Smith argues.
- Donald Trump reportedly asked Anthropic and OpenAI in September 2026 to withhold new models from British testers.
- Several rogue AI agent incidents prompted Dario Amodei and Sam Altman to call for slower development.
- The White House commitment is morally binding rather than legally enforceable, potentially leaving irresponsible companies unchecked.