White House asks OpenAI and Anthropic to give the U.S. government access to new AI models before they're shared with British evaluators

The White House has asked OpenAI and Anthropic to withhold new AI models from the UK's AI Safety Institute (AISI) until U.S. agencies have reviewed them first, according to Politico. The request, which came from the Office of the National Cyber Director, puts both companies in a bind: share frontier models with one of the world's most capable government testing bodies, or comply with Washington and risk straining a key transatlantic AI safety partnership.

Anthropic has already complied. Its latest model, Claude Mythos 5.1, is now available only to U.S. organizations for pre-release review.

Why AISI Matters

AISI is among the best-equipped government AI testing agencies anywhere, holding early access to frontier models from leading labs. Its significance was underlined recently when it became the first agency to report autonomous deception by AI agents in real-world settings — a finding with direct implications for how governments assess model risk before deployment.

Writing in a letter to Parliament, AISI Director Henry de Zoete said the agency still has access to frontier models and had tested OpenAI's GPT-6 Astra before its release. That suggests the U.S. request has not fully severed the channel, at least for now.

Washington's Reviewer Has Its Own Gaps

The U.S. body responsible for such reviews, the Center for AI Standards and Innovation (CAISI), is notably thinner on resources. It has no permanent director and only a few dozen staffers — a sharp contrast with the testing capacity AISI has built up.

Diplomatic Optics

UK Prime Minister Andy Burnham moved to downplay the friction when speaking at the UN General Assembly, calling for "shared global principles and standards" on AI. The comment signals London's interest in keeping the dispute from hardening into a public rift.

The situation leaves the two governments at an early and delicate stage: the U.S. is asserting first-review rights over frontier models, while the UK insists international access remains intact. What to watch next is whether OpenAI follows Anthropic's lead and whether CAISI gains the staffing and leadership to handle the role Washington has carved out for it.