The Sheffield Press

World

UK seeks deeper AI safety access from OpenAI and Anthropic

By Sarah Mitchell ·
UK seeks deeper AI safety access from OpenAI and Anthropic

The UK AI Security Institute had in-depth access to Anthropic and OpenAI during frontier model evaluations, including non-public tooling and safeguard details. The public record still does not show who inside government signed off on safety or what evidence carried the most weight.

Britain laid the groundwork on October 25, 2023, when the government published its frontier AI discussion paper and warned of significant uncertainty around the capabilities and risks of advanced models. By the AI Seoul Summit in 2024, the UK and Republic of Korea had announced Frontier AI Safety Commitments with major labs, including OpenAI and Anthropic, turning those concerns into a voluntary framework rather than a binding licensing regime.

OpenAI's Preparedness Framework uses a cross-functional Safety Advisory Group to review model capability reports and recommend safeguards before deployment. Only models with a post-mitigation score of high or below can be developed further. The record does not show whether British officials relied on company-run testing, independent testing from the AI Security Institute, or a blend of both.

AI-generated illustration
AI-generated illustration

The AI Security Institute is a directorate of the Department for Science, Innovation and Technology. The visible record still leaves unresolved which officials reviewed the evidence, how much distance they kept from the firms under scrutiny, and where the final judgment on safety was made.

The government's relationship with OpenAI was formalized further in 2024, when OpenAI CEO Sam Altman and the UK Secretary of State for Science, Innovation and Technology, Peter Kyle MP, signed a memorandum of understanding focused on AI adoption, infrastructure development and technical information exchange. The agreement also covered expanding the UK’s sovereign AI capability. OpenAI and Anthropic are also both members of the Frontier Model Forum, the industry body launched in July 2023 with Google and Microsoft to promote safe and responsible frontier AI development.

worldOpenAIAnthropic