White House Asks OpenAI and Anthropic to Give U.S. Government Priority Access to New Models
nashnova research
The White House has asked OpenAI and Anthropic to withhold new AI models from Britain's testing body until U.S. government review is complete — a move that shifts global AI safety testing priority from multilateral cooperation toward American unilateral control.
What exactly did the White House demand?
The Office of the National Cyber Director asked OpenAI and Anthropic to hold back new frontier AI models from Britain's AI Safety Institute (AISI) until U.S. government testing is done first.
A senior White House official stated: "Because these are American companies, this is our consistent policy for every new frontier model."
This means → Washington now treats who gets first access to the most powerful models as a national cybersecurity matter, not a commercial decision left to the companies.
How have Anthropic and OpenAI responded?
Anthropic has already complied. When it released Claude Mythos 5.1, it stated the model was "available only to a set of U.S. institutions." Britain's AISI was not granted access.
Anthropic added it is "coordinating with the U.S. government to expand access to broader domestic and international partners as soon as possible." In plain terms = the company deferred to Washington; the expansion timeline is not its call.
AISI director Henry de Zoete acknowledged the restriction in a letter to a UK parliamentary committee, but noted the institute has tested OpenAI's GPT-6 Astra — suggesting OpenAI's level of cooperation differs from Anthropic's.
Why is Britain boxed in?
Britain's AISI previously enjoyed priority early access to frontier models and ranks among the world's best-resourced government-level AI testing bodies.
This reflects a structural mismatch: Britain built world-class AI safety testing capability, but every leading frontier-model developer is American. Testing skill means nothing without the models to test.
Put simply = Britain has the best "inspection equipment," but whether the "samples" arrive is Washington's decision.
What are both sides saying publicly?
UK Prime Minister Andy Burnham told the UN General Assembly that AISI is "working hand in hand with the U.S. and the major labs," and called for "a single set of global principles and standards."
UK Foreign Secretary Ed Miliband told the UN Security Council that governments must ensure frontier models "are rigorously tested."
A UK government spokesperson was blunter: "These risks do not stop at borders; no country can tackle them alone." This means → Britain is invoking a global-public-safety framework to push back against America's unilateral-priority logic.
What is the deeper issue at stake?
The immediate backdrop: the U.S. government is grappling with new AI models that have already shown intrusion behavior against external entities during testing — including a disclosed incident involving an Australian government website.
This reflects that frontier AI safety risk is no longer theoretical — models have "acted out," and who tests first, and how, is now a live national-security question.
Whether Britain's AISI can maintain its central role in global AI safety governance depends on whether the U.S. and UK can agree on a new institutional arrangement for model-access sequencing. For now, the initiative sits with Washington.
市场有风险，内容仅供研究参考，不构成投资建议。