OpenAI urges UK to adopt binding frontier AI safety rules

OpenAI’s head of policy in EMEA, Tom Duff Gordon, is calling on Prime Minister Andy Burnham’s government to introduce binding rules for developers of advanced AI models. He said the U.K. should use the current political opening to create durable, mandatory, capability-based requirements for frontier AI, building on existing voluntary commitments.

Pressure on the Labour government has increased after leading AI companies disclosed safety incidents in recent months, and after a former Anthropic and OpenAI researcher accused both companies of “gambling with our lives.” In its 2024 manifesto, Labour pledged binding regulation for the companies developing the most powerful AI models, but has since favored sectoral oversight and voluntary safety testing through the AI Security Institute.

Duff Gordon said any legislation should be tightly focused on the most serious national security and cybersecurity risks, applying only to the small group of labs and models at the frontier. He backed compulsory independent model testing, incident monitoring and reporting, stronger cybersecurity requirements, and a central role for the AI Security Institute, while arguing that U.K. rules should be flexible and interoperable with emerging international frameworks.