As of 12 October 2026, American AI developers can give Britain’s AI Security Institute access to models before release after the Trump administration reversed a request to hold them back, Reuters reported, citing the Telegraph. The earlier request had covered developers including OpenAI and Anthropic. The reversal gives those companies room to cooperate with British safety testers without falling foul of the administration, according to the report.
Key points
- The administration has reversed its request that developers keep new models from Britain’s AI Security Institute until release, according to the Telegraph.
- The British institute tests advanced models for their capabilities and potential misuse.
- On 9 October, a US AI task force said developers must report security incidents involving their models.
OpenAI and Anthropic can share models early
The change concerns access before a model is released. Under the earlier US request, developers such as OpenAI and Anthropic had been asked to keep new models from the British institute at that stage. The Telegraph reported that the administration has withdrawn that request. American developers can now provide early access to the British government for safety testing without running afoul of the administration.
AI Affairs previously reported the White House’s request to OpenAI and Anthropic to hold models from UK testers. The reversal changes the administration’s position on that early access. It does not turn participation in the British testing programme into a US requirement: the reported decision removes a request to withhold models rather than directing developers to submit them.
Reuters reported that spokespeople for Trump, OpenAI and Anthropic did not immediately respond to its requests for comment. The account of the reversal therefore rests on the Telegraph’s reporting, rather than a statement from any of those parties.
Britain’s AI Security Institute may receive early access
Britain established the AI Security Institute after the AI safety summit in 2023. Its work includes testing advanced models to assess what they can do and how they might be misused. Early access lets that work take place before a model reaches the public, which is the point at which the US request had constrained developers’ cooperation with the institute.
For the British government, the practical gain reported by the Telegraph is the ability to receive models from willing American developers ahead of release. For those developers, the change removes an objection from their own government to sharing models with the institute. The testing itself remains British work; the reported reversal concerns what US companies may give its testers.
US task force calls for incident reports
The reversal comes against a change in the administration’s approach to AI safety requirements. Trump has repeatedly dismissed calls for binding requirements on developers, Reuters reported. Allowing pre-release testing in Britain gives developers more latitude to take part in an international safety effort, while leaving the decision to share a model with them.
The administration’s newly launched AI task force said on 9 October that developers must report security incidents involving their AI models. That instruction concerns security incidents involving developers’ AI models, rather than access for British testers. It places a reporting obligation alongside the administration’s decision to allow voluntary pre-release cooperation with the AI Security Institute.
.jpg){kind=link}