US requests AI firms withhold new models from UK safety institute

The United States Office of the National Cyber Director has requested that AI developers OpenAI and Anthropic withhold their newest frontier models from the United Kingdom’s AI Safety Institute (AISI) until a US-led security review is completed.

Anthropic has already complied with the directive, restricting access to its Claude Mythos 5.1 model to a specific group of US-based organizations. Henry de Zoete, the director of the AISI, confirmed in a letter to a UK parliamentary committee that the institute lacked access to the latest Anthropic model, though he noted the institute had previously evaluated OpenAI’s GPT-6 Astra before its public release.

The move places AI companies in a position where they must choose between providing early access to the well-resourced UK testing agency or risking friction with the US administration. This development follows recent statements from the White House rejecting globalist schemes to control artificial intelligence.