UK AI Security Institute regains US model access: independent testing must not depend on diplomacy

The United States has reversed a request that kept the UK's AI Security Institute (AISI) from testing new frontier AI models before release, The Telegraph reported on 11 October 2026, with LBC and other outlets following on 12 October. Politico reported on 24 September that the White House's Office of the National Cyber Director had asked OpenAI and Anthropic to hold new models back from UK testers until US authorities had reviewed them. Anthropic did not give the institute its Claude Mythos 5.1 model, and AISI director Henry de Zoete has said nobody outside the US had seen Anthropic's latest model before launch, as reported by The Guardian. The reports say UK AI minister Kanishka Narayan lobbied for the change. At the time of reporting, neither the White House nor the companies had commented on it.

The institute was set up after the first international AI Safety Summit in 2023 to assess advanced AI systems for their capabilities and potential for misuse. Its pre-release tests are among the few independent checks governments get on systems that millions of pupils, parents and workers will then use. Since late September, according to the reports, the people whose job is to run those tests have been asked to wait while US authorities carried out their own review. That makes this a story about trust, not just diplomacy.

Dan Fitzpatrick, The AI Educator, says the episode matters well beyond Whitehall. "Every school leader and employer is being asked to take a view on whether AI tools are safe enough to use," he says. "The honest answer is that most of us are relying on what the companies tell us. Independent testing before release is how that changes, and it should not hinge on a diplomatic favour between two governments."

His position is that the answer is neither alarm nor waiting. "I am not arguing for fear. Curiosity over fear has always been my starting point, and teachers and workers are already using these tools, often well ahead of policy. But curiosity includes asking for evidence. Outsource the doing, not the thinking, and that has to include the thinking about who has checked the thing you are handing the work to."

For school leaders, educators, parents and business leaders, Fitzpatrick suggests four steps. First, ask every AI vendor you use what independent testing its product has had, by whom, and when, and keep the answer on file. Second, do not treat a company's own safety statement as the whole picture; look for a regulator, researcher or public body that has examined it. Third, keep a human accountable for any decision an AI tool influences, from marking to hiring. Fourth, teach pupils and staff to ask the same question, because knowing how to question an AI system is now a core part of AI literacy.

Fitzpatrick works with schools, education bodies, governments and employers on adopting AI with people at the centre, and writes regularly on what AI means for learning, work and leadership. The question behind this story, how much any of us should trust AI we cannot check, is the one he is asked most often by school leaders and parents.

Dan Fitzpatrick is available for interview and comment on this story. He is available today for live or pre-recorded radio, TV, podcast and print interviews, with a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io