White House tells US labs to let it vet new models before UK safety testers do
Made public on 24 September, the policy was already in force: Mythos 5.1 had reached no non-US organisation since launch; by 12 October AISI was testing Gemini 4 Argon pre-release.
- Government & policy
- Safety & alignment
- Notable
The White House’s Office of the National Cyber Director asked OpenAI and Anthropic to withhold new models from the UK’s AI Security Institute (AISI) until US agencies had reviewed them, Politico reported on 24 September. City A.M. quoted a senior US administration official as saying this was “our policy with every new frontier model that comes out”, because the companies were American.
The report made public an arrangement that was already operating rather than a new order. Claude Mythos 5.1, released on 1 September, had been available only to US organisations from launch. AISI’s director, Henry de Zoete, said so in a letter to the Commons business committee dated 15 September, nine days before the Politico story. The letter did not mention any White House request. It said AISI kept “long-standing arrangements” with developers and had tested OpenAI’s GPT-6 Astra before its release that month. Anthropic told City A.M. it was “coordinating with the US government to expand access to a broader set of domestic and international partners as quickly as possible.”
The restriction followed earlier US intervention over Mythos. In May the administration had blocked Anthropic from widening access to Mythos Preview and said it was weighing review of models before release.
The Telegraph reported on 11 October that the White House had lifted the restriction, according to the Decoder’s account. Some coverage described this as the end of a 17-day freeze, counting from the Politico report. On the record above, the limits on Mythos 5.1 had begun at least three weeks before that report. UK AI minister Kanishka Narayan said in a speech the next day that the government had spent “the last few weeks” working with “the US government, the labs” to “retain access for Britain’s AI Security Institute to the most powerful AI models before they are released”. He said AISI was now evaluating the latest models, among them Google DeepMind’s Gemini 4 Argon. His letter to the same committee credited “recent international diplomacy”. Neither named the restriction.
AISI’s pre-release access rests on voluntary arrangements with the labs, not on any binding agreement. Neither the request nor its reversal was announced by the US government; both became public through press reporting and UK parliamentary correspondence.
In the commentary
What people were saying around this time — external links, from the record's commentary rail.
- 24 September 2026 · Shakeel Hashim · TransformerHacking is the least worrying part of OpenAI’s Australia incident
- 25 September 2026 · Zvi Mowshowitz · Don't Worry About the VaseOn Ezra Klein’s Podcast With Jensen Huang
- 23 September 2026 · Dylan Xu, Sebastian Prasanna, Alek Westover · Redwood ResearchAstra is much better at reasoning with filler tokens than previous models