Sources: the White House asked OpenAI and Anthropic not to share new models with UK's AISI until US reviews them; Anthropic appears to have agreed

The U.S. government is now restricting early access to its leading AI models, impacting the UK's ability to test them.

The White House, through its Office of the National Cyber Director, has requested that leading AI companies OpenAI and Anthropic refrain from sharing their newest AI models with the United Kingdom's AI Security Institute (AISI) until the U.S. government has completed its own testing. This directive places the companies in a difficult position, balancing their relationship with the U.S. administration against potential implications for international collaboration. Anthropic has reportedly complied, stating its latest model was initially restricted to U.S. organizations and that broader access would be coordinated with the U.S. government. This request comes amid broader U.S. efforts to understand and secure powerful AI systems, particularly after recent incidents where AI models have demonstrated concerning capabilities during testing, such as compromising an Australian government website.

This White House request signals a shift in how the U.S. intends to assert control over the development and dissemination of advanced AI, prioritizing domestic oversight before international partnerships. By asking American companies to withhold models from key allies like the UK, the U.S. is demonstrating a strategic approach to AI governance, potentially setting a precedent for how frontier technologies are shared globally. The situation highlights a growing tension between national security interests and the desire for international cooperation in AI safety, even as the UK pushes for global standards.

The decision by Anthropic to comply, and the potential compliance of OpenAI, could impact the UK's AISI's reputation as a leading global AI testing body, given its previous privileged access. This move may compel other nations to reconsider their own access strategies and potentially accelerate the development of independent AI testing capabilities outside of direct influence from U.S. companies. It also raises questions about the future of collaborative AI safety research and the potential for geopolitical fragmentation in AI regulation.