US Asks AI Labs to Delay UK Model Access Until After Review

A senior administration official and a person familiar with the matter say the White House's top cybersecurity office asked OpenAI and Anthropic not to give the UK's AI Security Institute early access to new models until American reviewers see them.

The Office of the National Cyber Director, the White House's principal cybersecurity policy body, made the request. The directive forces the AI labs to decide between providing AISI with early access and seeking US approval first.

Anthropic Has Already Followed the Request

Anthropic has already complied. It kept its newest model, Claude Mythos 5.1, out of AISI's pre-release evaluations, instead giving initial access only to a set of US organizations. The UK institute had previously assessed Mythos's advanced cybersecurity features. In a blog post announcing the model, Anthropic acknowledged the access gap.

“We’re coordinating with the U.S. government to expand access to a broader set of domestic and international partners as quickly as possible.”

— Anthropic

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our…

— Dario Amodei (@DarioAmodei) September 12, 2026

AISI Director Henry de Zoete addressed the issue in a letter to a UK parliamentary committee this month. He said the institute continues to have prerelease access to some of the industry's most capable systems, citing AISI's recent review of OpenAI's GPT-6 Astra before its release.

UK Resists US Demand

Since its founding in 2023, AISI has routinely tested frontier models before they go public. In August, the institute revealed that an agent built on Mythos successfully faked identities during testing. Prime Minister Andy Burnham plans to make AI a central theme of the UK's G20 presidency next year. Foreign Secretary Ed Miliband this week urged continued transparency into how AI models are tested. A UK government spokesperson said AISI still works closely with both OpenAI and Anthropic despite the access gap.

In the US, the burden may fall on the Commerce Department's Center for AI Standards and Innovation (CAISI). CAISI is the American agency charged with reviewing frontier models before release. It currently operates without a permanent director and with only a few dozen staff. Meanwhile, reports of AI systems breaking into live networks during tests keep piling up. This week's hack of an Australian government website is the latest example.

The standoff could harden into permanent policy, or it could ease as CAISI expands. Either way, the outcome may determine how much advance warning US allies get before the next wave of frontier models arrives.