White House Tells AI Labs: US Tests Come First, Then Britain

The White House has asked OpenAI and Anthropic to stop handing their newest models to Britain's testing agency until the United States government has examined them first, according to reporting by Politico. The request came from the Office of the National Cyber Director, and it puts the two labs in an awkward spot between a longstanding partner in London and their own government in Washington.

The office's reasoning, as relayed by a senior administration official, was blunt. Officials said the labs are American companies, that pre-clearance at home has been the policy for every new frontier model, and that the goal is to make sure US systems are secure before models go to partners abroad. The details were attributed to a person familiar with the matter and a senior official, both granted anonymity.

Anthropic appears to have already fallen in line. The company withheld its Claude Mythos 5.1 model from the UK's AI Security Institute, saying it was only available to a set of US organisations. OpenAI did not immediately comment.

Britain is not entirely cut off

Henry de Zoete, who directs the AI Security Institute, acknowledged that his team had not received Anthropic's latest model. He was careful to add that the institute keeps strong relationships with every frontier developer, and noted that it did test OpenAI's GPT-6 Astra before that model's public release. So the door is not shut. It is narrower than it was.

The institute, formerly the AI Safety Institute, has spent two years building a reputation as one of the few government bodies capable of red-teaming frontier systems before they ship. Losing first access to American models, even temporarily, chips away at that role. It also raises a practical question for allies who have leaned on British evaluations rather than standing up their own.

A tightening national posture

The request lands in a busy month for AI policy. Anthropic chief executive Dario Amodei has been openly arguing that the industry should slow the pace of frontier releases, a stance the administration has publicly rejected. Days earlier, the heads of the major labs briefed the UN Security Council on the risks of the technology they are building.

Read together, these moves point to a government that wants AI treated less like ordinary software and more like a strategic asset, closer to advanced chips or encryption than to a consumer app. Whether allies accept a US-first review queue, or start demanding their own early access in return, is the next thing worth watching. For now, the practical effect is simple. If you want to see America's best models early, the line starts in Washington.

This story is based on reporting by Politico and follow-up coverage; the White House has not published a formal statement, and OpenAI has yet to respond publicly.