The Pulse
White House Reverses Request to Restrict UK Access to AI Models
The Trump administration reversed a request that had asked OpenAI and Anthropic not to share new AI models with the UK’s AI Security Institute before release. The change restores US developers’ ability to provide British testers with early
AI.info Team · Written with AI ·
“No country can tackle the risks presented by frontier AI alone, and they do not stop at national borders.”
Henry de Zoete, director of the UK AI Security Institute
Henry de Zoete wrote those words to a UK parliamentary committee on September 15. Less than a month later, the White House reversed a request that had asked US developers not to share new AI models with Britain’s AI Security Institute (AISI) before release, according to a Telegraph report carried by Reuters on October 11.
The change means American developers can again give the British government early access to models for safety testing without running afoul of the Trump administration, Reuters reported. The wire service named OpenAI and Anthropic as developers affected by the earlier request. It did not specify when the restored access takes effect or whether new conditions apply.
A reversal after a dispute over model access
Reuters described the original measure as a decision requesting developers such as OpenAI and Anthropic not to share new models with AISI before release. The report attributes both the restriction and its reversal to The Telegraph; it does not identify a public White House order or provide the terms of the request.
That distinction matters. The reported shift gives developers greater latitude to work with British testers, but the account does not establish a formal guarantee of continuing access. Reuters said representatives for Donald Trump, OpenAI and Anthropic had not immediately responded to requests for comment.
Why pre-release testing matters to AISI
AISI evaluates advanced models to understand their capabilities and potential for misuse. The institute has said its work includes testing for cyber, chemical and biological risks, as well as autonomous capabilities and safeguards. Its first-year account says OpenAI, Google DeepMind and Anthropic have worked with the institute to test their most advanced models.
Pre-release access gives evaluators a chance to examine a system before it becomes publicly available. AISI’s parliamentary letter also makes clear that such access is not universal: de Zoete wrote that the institute continued to receive some pre-release models, and cited its testing of OpenAI’s GPT-6 Astra before public release.
Anthropic’s Mythos 5.1 showed the access gap
In the same letter, de Zoete said Anthropic had made clear that no organizations outside the United States had access to Claude Mythos 5.1 when it was released. He described AISI as being in daily contact with industry partners, including Anthropic, while acknowledging that the institute had not tested that model before release.
The contrast in the letter is concrete: AISI had tested GPT-6 Astra before launch, but did not have Mythos 5.1. Reuters’ account of the reversal does not say whether British officials will now receive Mythos 5.1, nor does it spell out which future releases fall within the restored access.
Access remains the unresolved detail
The White House’s reversal changes what US developers may do, but the public account leaves the practical terms open. It does not say whether US agencies must review models first, how early British testers will receive access, or whether every developer is covered.
For now, the clearest measure of the change is the one Reuters reports: US developers can again provide Britain with early access for safety testing. AISI’s letter records the recent exception that made that access issue tangible: it had tested GPT-6 Astra before release, while Anthropic’s Mythos 5.1 was unavailable outside the United States.