Washington drops the UK testing ban: OpenAI and Anthropic can share frontier models with Britain's AI Safety Institute
The Trump administration has reversed its earlier request that US labs keep new models from Britain's AI Security Institute until after release, the Telegraph reported Sunday — opening the door to early-access safety testing by London.
Washington has reversed course. Reuters reported on Sunday, citing the Telegraph, that the Trump administration dropped its earlier decision requesting that US developers such as OpenAI and Anthropic not share new AI models with Britain's AI Security Institute before release.
The practical meaning: American AI labs can now give the British government early access to frontier models for safety testing without running afoul of the White House. Reuters says spokespeople for Trump, OpenAI and Anthropic did not immediately respond to its request for comment.
From block to clearance in sixteen days
The reversal undoes a position we covered on September 25: the White House had told OpenAI and Anthropic to let Washington test new models before Britain's institute got access. That sequence — Washington first, London second — is now off the table.
Reuters frames the shift as a softer stance by the administration on international AI safety efforts, giving US developers more latitude to cooperate with Britain as policymakers work out how to oversee increasingly powerful systems.
Why London's testing role matters
Britain's AI Security Institute was created after the world's first AI Safety Summit at Bletchley Park in November 2023. Its job is to test the most advanced AI models for capabilities and misuse risk — ahead of public release.
Early access matters because frontier-model evaluations lose most of their bite when they happen after deployment. An institute that can probe a model before it ships can catch dangerous capabilities and jailbreaks while fixes are still cheap — exactly the kind of testing the earlier block had put on ice.
A shifting line, not a settled one
Reuters notes Trump has repeatedly dismissed calls for binding safety requirements on AI developers — a position that now appears to be moving. His newly launched AI task force said Friday that developers must report security incidents with their models.
What remains unknown: whether the clearance arrived with conditions, what "early access" actually covers — API access, evaluation harnesses, or full weights — and whether the same latitude extends beyond Britain. Sunday's report reopens the door; it doesn't describe the hallway behind it.