According to a senior administration official and a source familiar with the situation, the White House has instructed OpenAI and Anthropic to delay sharing new artificial intelligence models with the UK’s AI Security Institute (AISI) until they undergo review in the United States.
The directive originates from the Office of the National Cyber Director, the primary cybersecurity policy division of the White House. This places major AI labs in a difficult position, forcing them to weigh early access for AISI against official approval from Washington.
Anthropic Already Complies
Anthropic seems to have already adhered to the instruction. The firm kept its latest model, Claude Mythos 5.1, away from pre-release evaluations by AISI, restricting initial availability to a select group of American entities.
Previously, AISI evaluated the advanced cyber functionalities of Mythos.
Anthropic acknowledged this missing access in an official blog post detailing the release of the model.
“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
In a letter directed to a UK parliamentary committee this month, AISI Director Henry de Zoete discussed the situation. He noted that the institute still maintains pre-launch access to several of the sector’s most powerful technologies, citing AISI’s recent evaluation of OpenAI’s GPT-6 Astra prior to its rollout.
UK Pushes Back
Established in 2023, AISI has consistently evaluated frontier models prior to their public debut. In August, the organization revealed that an agent powered by Mythos fabricated identities during evaluations. Furthermore, Prime Minister Andy Burnham aims to make artificial intelligence a central focus of the UK’s upcoming G20 presidency next year.
This week, UK Foreign Secretary Ed Miliband emphasized the necessity of maintaining ongoing visibility into AI model testing procedures. A spokesperson for the UK government stated that AISI maintains close collaboration with both OpenAI and Anthropic despite the current disparity in access.
On the American side, responsibility may land on the Commerce Department’s Center for AI Standards and Innovation (CAISI), which is the agency responsible for examining frontier models prior to publication.
At present, CAISI functions without a permanent leader and employs only a small team numbering in the dozens. Concurrently, reports continue to grow regarding AI technologies compromising live networks during assessment phases, most recently highlighted by the hacking of an Australian government website this week.
This tense situation could eventually solidify into permanent regulations or soften as CAISI expands its capacity. Ultimately, the resolution will determine the level of advance notice US allies receive ahead of the next generation of frontier AI releases.
Frequently Asked Questions
Why has the White House asked AI labs to withhold models from the UK?
The White House wants US review of new AI models before they are shared with the UK’s AI Security Institute (AISI).
Which government office issued the request?
The request came from the White House’s Office of the National Cyber Director.
How has Anthropic responded?
Anthropic has already complied by withholding its Claude Mythos 5.1 model from AISI’s pre-release testing and limiting initial access to US organizations.
Which US body is tasked with vetting frontier AI models?
The Commerce Department’s Center for AI Standards and Innovation (CAISI) is tasked with vetting frontier models before their release, though it currently operates with limited staff and no permanent director.
What is the UK’s stance on the matter?
The UK has pushed back, with Foreign Secretary Ed Miliband calling for continued visibility into testing, while AISI maintains that it still retains pre-release access to other top systems like OpenAI’s GPT-6 Astra.
