The White House may be asking OpenAI and Anthropic to let the US government test new models before sharing them with the UK’s AI Security Institute. That request has not been independently verified, leaving the status and scope of any US-first testing requirement uncertain.
If applied, the arrangement would put British evaluators behind their US counterparts in the testing sequence. For UK AI model testing, access would depend on when Washington finished its review.
Anthropic describes a restricted rollout
Anthropic says access to Claude Mythos 5.1 is limited to a selected group of US organizations. The company says it is coordinating with the US government to expand availability to more domestic and international partners.
The connection between those access limits and any White House request remains unconfirmed. Restricted availability alone does not establish why the UK institute lacked access or whether the same conditions would apply to OpenAI.
A firm timetable for access by the UK institute has not been publicly confirmed.
The UK describes uneven access
In a letter to a parliamentary committee, AISI director Henry de Zoete acknowledged the lack of access to Anthropic’s model. He said the institute had tested OpenAI’s GPT-6 Astra before its public release and retained prerelease access to some of the most capable models.
His account suggests access can differ by developer and release. It also leaves open how a US-first review process would affect future evaluations, even where the institute has an established relationship with a developer.
Cooperation hinges on when models arrive
The UK government says the institute continues to work closely with the US government, OpenAI and Anthropic. Its position is that AI risks cross national borders and require cooperation between countries.
The practical issue is sequencing. If British evaluators must wait for a US review to finish, their opportunity to conduct prerelease AI safety testing could narrow. The effect would depend on the length of the US review and the developer’s release schedule.
The key unanswered questions concern which models would be covered, how long reviews would take and when UK access would follow. Those details would determine how far such an arrangement changes international AI testing.