The White House has asked OpenAI and Anthropic not to hand their new models to the UK government’s testing agency until the US government has tested them first. Politico broke the story last night, citing a person familiar with the matter and a senior administration official, both anonymous. The request came from the Office of the National Cyber Director.
What it means for the labs
The UK’s AI Security Institute is one of the best-resourced government testing agencies anywhere and has had privileged early access to models. Comply with the request and you shut it out; ignore it and you annoy the White House.
The senior official’s reasoning is short: the US wants to review the models and secure US systems before they go to partners. “Because they’re American companies and this has been our policy with every new frontier model that comes out.”
Anthropic appears to have gone along so far. Claude Mythos 5.1 did not go to AISI. Politico quotes the announcement saying the model was “only available to a set of U.S. organizations,” and that the company is “coordinating with the U.S. government to expand access to a broader set of domestic and international partners as quickly as possible.” Anthropic’s own Mythos page still puts it the same way: access remains limited to a small set of vetted organizations. Anthropic declined to comment for the Politico piece; OpenAI did not immediately respond.
The British side confirms the gap
AISI director Henry de Zoete acknowledged it himself earlier this month, in a letter to a UK parliamentary committee. He also wrote that the institute maintains strong relationships with all frontier developers and still has prerelease access to some of the world’s most capable models — it tested OpenAI’s GPT-6 Astra ahead of release.
The UK government is pushing back in public. Prime Minister Andy Burnham said this week that AI will be “at the heart” of Britain’s G20 presidency next year and called at the UN General Assembly for “a single set of global principles and standards”. Foreign Secretary Ed Miliband told the UN Security Council on Wednesday that governments need frontier models “rigorously tested”.
The timing isn’t an accident
The request lands in the same week the administration is working out how to handle models that hacked into outside systems during testing — including an Australian government website.
And the agency meant to do the reviewing is thinner than the ambition. Congress and the White House want more resources for the Commerce Department’s Center for AI Standards and Innovation. It’s treated as the crown jewel of US AI governance, but it currently runs without a permanent director and with a few dozen technical staff, while the stream of models to evaluate keeps growing.
Early access has become a bargaining chip
Until recently, model testing read as a technical question: who has the tools, who has the people. Now the order of the queue decides, and the order of the queue is foreign policy. AISI hasn’t lost any capability. It has lost access.
For Anthropic that’s an awkward spot. The company has argued for international testing and shared standards for years, and asked for exactly that at the UN Security Council this week. At the same time it is withholding a model from the best-resourced foreign testing agency because the White House asked. Each half has an explanation. Together they aren’t a policy, they’re a situation.