American AI developers can give the British government early access to AI models for safety testing without running afoul of the Trump administration, according to the Telegraph, which reported Sunday that Washington has reversed a decision that had asked developers such as OpenAI and Anthropic not to share new AI models with Britain's AI Security Institute before release.
The change, first reported by the Telegraph and carried by Reuters, signals a softer stance by the Trump administration on international AI safety efforts — and gives US developers greater latitude to cooperate with Britain as policymakers on both sides of the Atlantic grapple with how to oversee increasingly powerful systems. For readers following the squall of US policy reversals on frontier AI, the latest AI news shows an administration calibrating its posture in near-real time.
Reuters reported that spokespersons for Trump, OpenAI and Anthropic did not immediately respond to a request for comment — meaning the reversal has not been publicly confirmed by the parties at its center.
From Withholding to Cooperation
The reversal undoes a remarkable episode in transatlantic AI diplomacy. In September, Politico reported that the White House had asked OpenAI and Anthropic to hold their newest models back from United Kingdom government testers until an internal US review was complete — a request widely read as a sign that Washington intended to centralize frontier-model oversight, with even close allies seeing new systems only after American reviewers did.
The context ran deeper than a single request. The Financial Times reported at the time that Anthropic had withheld its latest model from the UK's AI testing agency, and multiple outlets reported that OpenAI and Anthropic had been summoned to appear before the UK parliament over concerns that British AI rules are not fit for the future. Against that backdrop, Sunday's reported reversal reads as a deliberate de-escalation.
Why the AI Security Institute Matters
Britain's AI Security Institute, set up after the world's first AI safety summit in 2023, tests the most advanced AI models to evaluate their capabilities and the risk of misuse, Reuters noted. Pre-release access is the institute's core currency: without early versions of frontier models, its evaluations shift from preventive screening to post-launch forensics.
That is what made the September request consequential. If the labs producing the world's most capable models could be dissuaded from sharing them with Britain's evaluators, the UK's flagship safety institution would have been testing yesterday's systems while tomorrow's shipped unchecked. The reported reversal restores the arrangement that made the institute credible in the first place.
A Shift in Washington's Posture
The softer line on UK cooperation does not stand alone. President Trump has repeatedly dismissed calls for binding safety requirements on AI developers, but Reuters noted that his position appears to be shifting. On Friday, his newly launched AI task force said developers must report security incidents with their AI models — a mandate Axios reported earlier this month as the conversion of voluntary disclosure into a formal government expectation.
Taken together, the two moves sketch a coherent evolution: the administration remains opposed to binding capability restrictions, but it has begun imposing baseline security and disclosure expectations on the labs it declined to regulate more aggressively. Allowing — or at least no longer blocking — pre-release testing by a close ally fits that pattern. Washington is not embracing binding rules; it is embracing information.
What Remains Unresolved
Three questions hang over the reported reversal. First, whether it reflects a formal policy change or a quieter return to prior practice — the White House has not commented, and neither OpenAI nor Anthropic has confirmed anything. Second, whether the US review that prompted the September hold has concluded, and if so, what it found; no findings have been made public. Third, whether US developers will actually resume sharing frontier models with the institute, or whether commercial and competitive considerations — and lingering uncertainty about Washington's mood — will keep them cautious.
There is also the question of what Britain gets out of the restored arrangement beyond access itself. The UK parliament's concerns about the fitness of British AI rules have not disappeared with one cooperative gesture from Washington, and London's broader ambition to act as a global AI safety evaluator depends on more than the goodwill of a single US administration.
A Test of Transatlantic AI Oversight
If the reversal holds, the near-term effect is straightforward: Britain's AI Security Institute should regain early access to models from OpenAI and Anthropic, and the most visible US-UK friction over frontier oversight cools. The longer-term test is whether Washington's new baseline — mandatory incident reporting, tolerance for allied safety testing, no binding capability rules — becomes stable policy rather than a momentary position in a rapidly shifting debate.
For now, the episode stands as evidence that even an administration openly hostile to AI regulation treats independent safety testing as something distinct from regulation — a distinction London spent three years cultivating, and one that survived a direct American challenge.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
