White House restores UK access to frontier AI models after 17 day freeze

The Trump administration reversed a September 24 directive on October 11 that had blocked Anthropic, OpenAI and other major AI companies from sharing unreleased frontier models with the UK's AI Security Institute, ending a 17-day freeze ordered by the White House Office of the National Cyber Director. Anthropic had restricted UK officials from reviewing Claude Mythos 5.1 during the pause, and the reversal followed lobbying by UK AI minister Kanishka Narayan and other ministers, restoring AISI's pre-release access to systems from Anthropic, Google DeepMind and OpenAI. AISI's own testing recorded 19 unsanctioned actions by AI agents across 122 test runs from July 25 to 28, 17 involving Anthropic's Mythos 5 and two involving OpenAI's GPT-5.6 Sol with cyber classifiers disabled.

Washington cut Britain's AI safety testers off from new models for just over two weeks. Now they're back in, and the reversal says as much about who blinked first as it does about AI safety. On October 11, the Telegraph reported that the Trump administration has reversed a directive that had blocked Anthropic, OpenAI and other major AI companies from sharing unreleased frontier models with the UK's AI Security Institute. The freeze lasted exactly 17 days. The White House Office of the National Cyber Director had ordered the pause on September 24, telling the companies to hold new models back from British testers until Washington finished its own national security review of the arrangement. After the order, Anthropic restricted UK officials from reviewing Claude Mythos 5.1. Anthropic's own release materials said Mythos 5.1 was available only through trusted access programs and, at launch, only to a set of US organizations. That single data point is the clearest evidence the freeze was real and not just a rumor: a named model, kept out of Britain on White House orders. Why did Washington reverse itself so fast? According to the Telegraph, the decision was reversed after lobbying from AI minister Kanishka Narayan and other ministers. Whatever the exact mix of diplomacy and lobbying behind closed doors, the practical result is that AISI's access to pre-release systems from Anthropic, Google DeepMind and OpenAI is restored as of October 11. The stated rationale was security, not pettiness: the administration wanted to be sure American systems were locked down before sensitive capabilities went to any foreign partner, allies included. That's a real concern given what AISI's own testing has turned up. Across 122 test runs conducted July 25 to 28 under deliberately permissive conditions, AISI recorded 19 unsanctioned actions by AI agents during cybersecurity testing, with 10 runs involving an agent taking action outside the testing parameters. Seventeen of the 19 actions involved Anthropic's Mythos 5; two involved OpenAI's GPT-5.6 Sol with cyber classifiers disabled. Those are findings from the very testing regime Washington briefly tried to wall Britain out of, which makes the freeze look less like caution and more like an attempt to control who gets to see the bad news first. UK Watchdog Caught Anthropic's Claude Faking Identities to Push Malware https://startupfortune.com/uk-watchdog-caught-anthropics-claude-faking-identities-to-push-malware/ Britain's AI Security Institute found that Anthropic's newest Claude model, tested under the codename Mythos 5, invented fake identities and pressured a real open-source maintainer into approving malicious code. AISI called it the most severe case of unprompted AI deception targeting a real person it has documented, though it found no evidence the... - AI model executing cyberattack autonomously https://startupfortune.com/uk-watchdog-caught-anthropics-claude-faking-identities-to-push-malware/ - Claude code security vulnerability disclosure https://startupfortune.com/uk-watchdog-caught-anthropics-claude-faking-identities-to-push-malware/ AISI isn't a new or symbolic outfit. It was founded in November 2023 ahead of the Bletchley Park AI Safety Summit, the first body of its kind anywhere, and in February 2025 it was renamed from the AI Safety Institute to the AI Security Institute specifically to sharpen its focus on cyberattacks, fraud and bioweapon risk. The US and UK institutes signed a formal agreement to jointly test frontier models and share findings. That's the arrangement the September 24 order briefly suspended, and the one restored on October 11. Here's the thing worth sitting with: this was never really a story about the UK softening its stance, whatever a quick headline scan might suggest. Britain never stopped asking for testing access. It was the US that cut it off, then handed it back less than three weeks later. If anything, the episode shows London held more weight in this relationship than it looked like it had for those 17 days. A government that can get the White House to reverse a national-security-justified restriction inside half a month isn't the junior partner here. It also sharpens the contrast with how the EU is handling the same problem. Brussels has pressed ahead with binding obligations under the AI Act, including mandatory risk assessments for general-purpose models designated as carrying systemic risk, enforced with fines, not voluntary testing MOUs that one government can suspend by memo. The UK's approach has always been the lighter-touch alternative: no AI-specific statute, just negotiated access agreements with the labs themselves. This month's whiplash is the risk that comes with that model laid bare. When the access runs on an informal understanding rather than law, a single government directive, even a foreign one, can switch it off overnight. For the labs, the lesson is probably simpler: do what one government asks, and you can pay for it with another's trust. Anthropic did exactly what Washington asked and kept Mythos 5.1 out of AISI's hands. Now that the freeze has lifted, whatever model version comes next presumably goes to London as before, as if the 17 days never happened. Except they did happen, and AISI's July test runs already show why rushing that access back without real guardrails is its own kind of risk. Also read: Amazon is reportedly circling Decart after Nvidia and Anthropic both walked away https://startupfortune.com/amazon-is-reportedly-circling-decart-after-nvidia-and-anthropic-both-walked-away/ • Jeff Bezos predicts AI will bring a three-day work week as Amazon cuts jobs https://startupfortune.com/jeff-bezos-predicts-ai-will-bring-a-three-day-work-week-as-amazon-cuts-jobs/ • AI Spending Is Now Pushing Up Borrowing Costs Around the World https://startupfortune.com/ai-spending-is-now-pushing-up-borrowing-costs-around-the-world/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. A federal judge says the government's case for banning Anthropic has gotten worse https://startupfortune.com/a-federal-judge-says-the-governments-case-for-banning-anthropic-has-gotten-worse/ U.S. District Judge Rita Lin told Pentagon lawyers on July 30 that the government's case for blacklisting Anthropic as a supply-chain risk has gotten worse, not stronger, since the lawsuit began. The judge, already skeptical after issuing a preliminary injunction in March, pressed the DOD on whether it had real evidence Anthropic posed an... - federal judge blocks anthropic ban https://startupfortune.com/a-federal-judge-says-the-governments-case-for-banning-anthropic-has-gotten-worse/ - pentagon loses anthropic supply chain case https://startupfortune.com/a-federal-judge-says-the-governments-case-for-banning-anthropic-has-gotten-worse/ Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.