The White House asked labs to hold models from UK testers. Then it reportedly reversed course.
For about two and a half weeks, OpenAI and Anthropic were asked to keep new models from Britain's AI Security Institute. Anthropic complied. Then Washington reportedly changed its mind.
On September 24, Politico reported that the White House had asked OpenAI and Anthropic not to share their newest AI models with Britain’s AI Security Institute, the government team that tests frontier models before release. Seventeen days later, a Telegraph report carried by Reuters said the administration had reversed that request.
I couldn’t open the Telegraph original, and my tools couldn’t load the full Politico or Reuters pages. The quotes below come from search-result copies of those articles, so check the wording against the originals if you plan to quote them.
What did the White House ask for?
The request came from the Office of the National Cyber Director, according to Politico. It asked OpenAI and Anthropic to hold new models back from the AISI until the US government had finished its own testing. A senior administration official, speaking on condition of anonymity, gave one line as the reason: “Because they’re American companies and this has been our policy with every new frontier model that comes out.”
That’s a short reason for a rule that puts the US first in line. Politico described it as a request, not a regulation. I didn’t find any law or order behind it.
Why did Anthropic go first?
Anthropic released Claude Mythos 5.1 on September 1. The announcement said the model was “only available to a set of U.S. organizations,” and it went out through Project Glasswing, a restricted program for US-vetted defensive security work. Politico reported it as the first time the UK institute had been left out of a pre-release Anthropic evaluation.
Anthropic has said it is coordinating with the US government to widen access “to a broader set of domestic and international partners as quickly as possible.” No date came with that. OpenAI hadn’t said publicly whether it complied, as far as I could find.
What did the UK say?
The AISI’s director, Henry de Zoete, wrote to a UK parliamentary committee in mid-September. He acknowledged the Anthropic gap. He also said the institute “continue[s] to have prerelease access to some of the world’s most capable models,” and pointed to its pre-release testing of OpenAI’s GPT-6 Astra. A UK government spokesperson was blunter. “These risks do not stop at national borders and no country can tackle them alone.”
Read closely, de Zoete described a narrower channel, not a closed one. The word “some” is doing a lot of work.
Which models are missing? The reports don’t say. Nobody has said.
What changed on October 11?
The Telegraph reported, and Reuters carried, that the administration had reversed the request. American developers can now give the British government early access to models for safety testing without running afoul of the White House. Reuters said spokespersons for Trump, OpenAI and Anthropic hadn’t replied to requests for comment. I found no White House statement on the reversal.
So far this is one newspaper’s reporting, with no on-the-record confirmation I could find. Treat it as reported, not settled, until a company or the White House says so in public.
Does the reversal fix anything?
Not by itself. The reports don’t say why the request was dropped. Lasting about 17 days suggests it never had a firm footing. That’s my read, not a source’s claim.
The next model matters more than this headline. Anthropic said it would widen access to Mythos 5.1, and I couldn’t confirm that has happened. If OpenAI’s next frontier release reaches UK testers only after US reviewers have finished, the reversal was a pause. If UK testers get it on the old terms, that’s the real sign things are back to normal.
For what AISI found when it tested GPT-6 Astra, read the post on OpenAI cancelling GPT-6.1 Astra. For a different UK regulator’s move on AI companies, see the post on the ICO’s ten company commitments.
Sources
- Politico, “White House asks OpenAI and Anthropic to hold new models from UK testers until US review,” 24 September 2026: politico.com
- Reuters, “US reverses course on Britain’s testing of AI models, Telegraph says,” 11 October 2026 (syndicated copy): usnews.com