The White House told OpenAI and Anthropic to keep their newest AI away from Britain’s safety testers – and Anthropic already has

Britain’s AI safety testers are being frozen out of America’s newest models. The White House has asked OpenAI and Anthropic to keep their latest AI systems away from the UK’s AI Security Institute (AISI) until U.S. officials have finished their own review, Politico reports. Anthropic appears to be going along with it.

What the White House asked for

The request came on Thursday from the Office of the National Cyber Director. Washington wants to test new frontier models first and make sure U.S. systems are secure before partners get access, according to Politico. Asked why, a senior administration official gave a simple answer:

Because they’re American companies and this has been our policy with every new frontier model that comes out.

Senior Trump administration official, to Politico

The administration wants the Commerce Department’s Center for AI Standards and Innovation (CAISI) to be the main U.S. testing authority for frontier models.

Anthropic’s Mythos 5.1 skipped the UK

The first casualty is already visible. The UK institute didn’t get pre-release access to Anthropic’s Claude Mythos 5.1, AISI director Henry de Zoete has acknowledged. Anthropic has said the model is “only available to a set of U.S. organizations,” adding: “We’re coordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible.” It declined to comment further. OpenAI didn’t immediately respond.

AISI isn’t completely shut out. De Zoete pointed out that it tested OpenAI’s GPT-6 Astra before release this month, and said: “The trusted relationships AISI holds enable us to test a range of models that represent industry-wide capability jumps.”

‘No country can tackle them alone’

Pre-release testing is the whole point of the AISI, which Britain set up in 2023 and has become one of the best-known government AI testing teams in the world.

The UK government pushed back gently, Analytics India Magazine reports:

These risks do not stop at national borders and no country can tackle them alone.

UK government spokesperson

A busy week for AI safety politics

The timing is awkward. On Wednesday, Sam Altman and Dario Amodei told the UN Security Council that AI needs shared international testing, and the UK backed international frameworks in the same session. A day later, reports emerged that Google, OpenAI and Anthropic are planning their own industry safety body. Now the U.S. government is telling two of those labs to put America first on testing.

Washington’s caution isn’t coming from nowhere, though: officials are weighing how to handle models after tests found some could break into outside systems, including an Australian government portal.

Why it matters

If U.S. labs have to wait for Washington’s sign-off before sharing, Britain loses its early look at the models most likely to cause problems, which is exactly what its institute was built for. It also shows how hard the global testing the labs asked for at the UN will be to deliver: the first government to step in has put its own review ahead of its closest ally’s.

Sources: Politico, via Yahoo News, Analytics India Magazine