White House Restores UK Access to Pre-Release OpenAI and Anthropic Models After Cutting Them Off

Britain is back in the room. The UK's AI Security Institute will once again get to test the most powerful models from OpenAI and Anthropic before the public sees them, AI Minister Kanishka Narayan announced on Monday. The move reverses a US policy that had locked British testers out of the newest frontier systems for weeks.

The restored access goes beyond those two labs. Models from Google DeepMind are also back on the list, according to The Decoder, after The Telegraph first reported the White House had changed course.

How Britain got cut off

The cutoff came in September. The Trump administration told OpenAI and Anthropic to hold new models back from the UK institute until US authorities had cleared them first, Bloomberg reported. We covered that order when the White House first told the labs to keep new models from UK testers pending a US review.

The policy had real consequences. Anthropic stopped British officials from reviewing its Claude Mythos 5.1 model, according to LBC and Bloomberg. The company gave no explanation for withholding it.

That mattered because the UK institute is not a side project. Founded in 2023, it is the most generously resourced government body anywhere dedicated to assessing cutting-edge AI models. Pulling it out of pre-release testing meant one of the world's best-funded safety evaluators was suddenly looking at frontier systems from the outside.

Lobbying, executives and a King

Narayan credited a campaign of lobbying and recent conversations between British ministers and tech executives for the reversal. Some of those talks were helped along by a summit with King Charles III.

"I'm very confident that, as a consequence of the conversations we've had, we will be able to test the latest generation of models," Narayan told Bloomberg.

He was careful not to oversell it. "When you're reliant on someone else's technology, you have to be clear-eyed about the permanence of any outcome," he said.

That line carries the real story. Britain hosts one of the most capable AI testing operations in the world, yet the models it tests are built by American companies operating under American political direction. A single instruction from Washington was enough to shut the door, and the door reopened only after a round of diplomacy.

A fragile arrangement

Nothing in the announcement suggests the UK now tests models before the US does. The September order put American review first, and the reversal restores British access without detailing any new guarantees about timing or permanence.

For the labs, the episode exposed how quickly their safety commitments to foreign governments can be overridden at home. Anthropic withheld Mythos 5.1 from British reviewers without public comment, and nothing stops a similar instruction from arriving again.

Narayan is pressing ahead on his own track anyway. He plans to announce new AI safety measures, The Decoder reports, including stricter reporting requirements for AI-related security incidents that involve critical infrastructure.

For now, British testers get their early look back. How long it lasts depends on decisions made in Washington.

Some offers on this page may be paid placements or contain affiliate links.