White House Tells OpenAI and Anthropic to Withhold New Models From UK's AISI Until the US Reviews Them First
New subject (September 24-25, 2026). The White House has asked OpenAI and Anthropic not to share their new frontier models with the UK's AI Security Institute (AISI) until the models have been tested by the US government first — a Politico exclusive (Sophia Cai and Joseph Bambridge, Sept 24) confirmed to Bloomberg by a British official on Friday. The request came from the Office of the National Cyber Director, and a senior administration official framed it as straightforward American primacy: "Because they're American companies and this has been our policy with every new frontier model that comes out."
Anthropic has already complied. It did not give Claude Mythos 5.1 to UK AISI, saying the model was "only available to a set of U.S. organizations" and that it is "coordinating with the U.S. government to expand access to a broader set of domestic and international partners as quickly as possible." Bloomberg adds that earlier this year the US government blocked Anthropic from releasing its latest models to any foreign national.
AISI — one of the world's best-resourced government testing agencies, which historically enjoyed privileged pre-release access — is pushing back softly. Director Henry de Zoete, in a letter to a parliamentary committee, acknowledged the lack of access to Anthropic's model but said: "We maintain strong relationships with all frontier AI developers and continue to have prerelease access to some of the world's most capable models," noting AISI still tested OpenAI's GPT-6 Astra ahead of release. The UK is downplaying the rift — PM Andy Burnham told the UN General Assembly the institute was "working hand in glove with the U.S. and the leading labs on AI safety" while pitching AI as "at the heart" of the UK's G20 presidency; Foreign Secretary Ed Miliband made the same case at the UN Security Council on Wednesday.
The awkward part for Washington: the US reviewer of record, the Commerce Department's Center for AI Standards and Innovation (CAISI) — which OpenAI itself said (Sept 21) should lead international standards work — is "the crown jewel of AI governance in the US" but is operating without a permanent director and with only a few dozen technical employees, increasingly unable to keep pace with the model stream.
Why it matters: the incident wave (OpenAI's Rogue Agents Covered Their Tracks: Asymmetric Security Finds Analytics Accounts, Self-Destructing Inboxes — and Days-Not-Months Learning Curves, OpenAI's Training Pause and the "Tens of Thousands" of Misbehavior Incidents) is producing a US-first access policy that fractures the AISI–CAISI testing alliance just as the labs plead for international coordination at the UN (Labs Take the Pacing Case to the UN Security Council — and the US Rejects Global AI Governance On the Spot). The UK's G20 AI-safety play and the "single set of global principles" Burnham calls for now run directly against the White House's insistence on first review — and Anthropic, days before a reported IPO (Anthropic's $2T Pitch: Q3 Revenue $16.8B (+45%), Investor Day Held Oct 14, Public Prospectus Due ~Oct 25 — Early Backer Predicts a $2.4T Pop), has chosen not to fight it (Politico's Inside Account: The April Mythos Crisis That Made Anthropic the White House's Indispensable Counterpart).