The Trump administration has reversed a request that US AI developers keep new models from Britain’s AI Security Institute before release, Reuters reports, citing the Telegraph. If confirmed, the change gives developers more room to share models with UK safety testers before launch, a useful opening for international scrutiny rather than a guarantee of it.
Watch Desk analysis
What happened
Reuters reports that OpenAI and Anthropic are among the developers that can now give the institute early access without running afoul of the administration. The institute tests advanced models for their capabilities and risks of misuse. Reuters says the Telegraph reported the reversal on Sunday, 11 October.
The report describes a change in the administration’s position, not a new testing requirement. Spokespeople for the Trump administration, OpenAI and Anthropic did not immediately respond to Reuters’ requests for comment.
Why it matters
Pre-release access can give safety evaluators a chance to examine a model before it reaches the public. Allowing US developers to share models with a UK institute may widen that opportunity across borders, though access alone does not establish what testing will happen or what developers will do with the findings.
Our read
This is a meaningful shift in the reported direction of US policy, but the distinction between permission and practice matters. The useful test is whether developers actually provide early access, and whether evaluators can publish enough about their work to make that access count.
What to watch
- Whether the US administration or the developers confirm the reported change.
- Whether OpenAI and Anthropic provide pre-release access to the UK institute.
- What the institute can disclose about its evaluations and findings. A reported policy reversal could make it easier for Britain’s AI Security Institute to test US models before release. Reuters, citing the Telegraph, says developers including OpenAI and Anthropic can now share early access without violating the administration’s previous request. Permission is not the same as testing, of course. The key question is whether companies use the opening, and what the institute can share about its evaluations.
Discussion spark: Should pre-release AI safety testing across borders be a voluntary arrangement between developers and governments, or should access be required by law?
Sources and evidence
- US reverses course on Britain's testing of AI models, Telegraph says – Reuters (11 October 2026, 23:48 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.