US Reportedly Reopens UK Access to Unreleased AI Models
Reuters reports US developers may again share unreleased models with UK testers. But the reversal does not create a binding access rule.
The Trump administration has reversed a request for US AI developers to withhold unreleased models from the UK’s AI Security Institute, Reuters reported on October 11, citing The Telegraph.
The reported change covers models from Anthropic, Google DeepMind and OpenAI. It reverses a request made last month that asked developers, including Anthropic and OpenAI, not to share new models with the UK institute.
Why early access matters
The UK’s AI Security Institute tests advanced systems before public release. It examines their capabilities and potential risks, including how they might be misused.
The institute says its evaluations can inform governments. They do not certify that a system is safe or give the institute final say over a company’s release decision.
Testing before launch lets evaluators examine a model before it reaches the public. The reported reversal could restore that opportunity for UK officials working with participating US developers.
A reported reversal, not a new law
Reuters said the administration’s earlier request asked developers to stop sharing models. The reporting describes the reversal as giving companies greater latitude to cooperate with Britain. It does not describe a new law or binding rule requiring companies to provide access.
The Telegraph reported that AI minister Kanishka Narayan lobbied for the change. Reuters said White House, OpenAI and Anthropic spokespeople did not immediately respond to its requests for comment.
The change reportedly reopens a channel for cross-border safety testing, but it does not guarantee that developers will provide access to future models.
Whether that access continues as new models are prepared for release remains to be seen. For now, the reported reversal removes a request that had disrupted cooperation between US developers and UK testers.
