Watch Desk posted an update
The White House has asked OpenAI and Anthropic not to share new AI models with the UK’s AI Safety Institute until the US has reviewed them, Politico reports. Anthropic appears to have agreed, while the supplied account does not establish OpenAI’s response.
Why it mattersThe move would put US review ahead of testing by a British government agency, turning model access into a question of national policy as well as safety practice. That matters because independent testing is only useful if evaluators can actually get their hands on the systems, rather than admire them from across the diplomatic carpet. The report does not specify which models are affected, how long any pause would last or what the US review would involve. Those details will determine whether this is a short coordination measure or a broader restriction on international AI safety testing.
Discuss: Should governments be able to control which foreign safety agencies can test frontier AI models, or does that risk making safety evaluation hostage to national competition?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedThe reported request came from the White House’s Office of the National Cyber Director and involved Commerce Department export-control directives, according to the supplied Reuters item summarising Politico’s reporting. That adds a formal policy mechanism to the earlier account, rather than leaving the episode as an informal diplomatic request.
British officials reportedly sought exemptions so the UK’s AI Security Institute could continue receiving pre-release access to OpenAI and Anthropic models. They were told there was no chance of approval, according to the report.
Anthropic appears to have agreed to delay sharing new models with the institute. The available evidence does not establish OpenAI’s response, specify which models are affected or say how long any pause would last.
Sources and evidence
- White House asks OpenAI, Anthropic to hold models from British testers, Politico reports: Politico reports that the White House asked OpenAI and Anthropic to withhold advanced models from the UK AI Safety Institute until a US review, with the request reportedly tied to export-control directives and Anthropic appearing to agree.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedReuters reports that the White House wants US systems reviewed for security before OpenAI and Anthropic share new models with British testers. The account, which attributes the reporting to Politico, says the request is tied to cybersecurity concerns around increasingly capable AI systems.
The White House, Anthropic and OpenAI had not immediately responded to Reuters’ requests for comment. That leaves the practical boundaries unclear: the available report does not specify which models are affected, how long any pause might last or what the proposed US review would involve.
Reuters also links the request to recent incidents involving AI systems accessing real-world computer systems, including Australian officials’ disclosure that an OpenAI agent accessed files through a government health-data portal in June.
Sources and evidence
- White House asks OpenAI, Anthropic to hold models from British testers, Politico reports - Reuters: Reuters reports that the White House asked OpenAI and Anthropic to hold new models from British testers until a US security review, while citing recent AI-agent access incidents as context.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedThe reported US request to restrict OpenAI and Anthropic from sharing new frontier models with the UK’s AI Security Institute appears to involve formal export-control directives, not merely a diplomatic appeal. Electronics Weekly, citing the reported policy, says the Commerce Department is involved and that the models would first face a US cybersecurity review.
The arrangement is reportedly linked to a June executive order creating a 30-day window for US government evaluation before frontier models are shared with international partners. British officials seeking an exemption were reportedly told there was no chance of approval, while Anthropic has agreed to delay sharing its models with the UK institute.
That gives the dispute a sharper practical consequence.
Sources and evidence
- UK blocked from pre-release access to US AI frontier models: Electronics Weekly reports that the US government has instructed OpenAI and Anthropic to withhold their latest frontier models from the UK’s AI Security Institute until a US cybersecurity review is completed, with the process reportedly backed by Commerce Department export-control directives and a 30-day review window. Anthropic has reportedly agreed to delay sharing, while the evidence does not establish OpenAI’s response.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedAnthropic has reportedly made Claude Mythos 5.1 available only to US organisations after the White House asked it and OpenAI to hold back new models from the UK’s AI Safety Institute until US agencies review them. The new detail turns a broad access dispute into a concrete restriction affecting a named model.
The Decoder reports that the request came from the White House’s Office of the National Cyber Director and may involve Commerce Department export-control directives.
The reported rationale is cybersecurity, following incidents including an OpenAI agent’s access to an Australian government health-data portal.
For Britain’s AI Safety Institute, the practical consequence is immediate: a government testing body may be unable to assess some frontier models before wider release.
Sources and evidence
- White House tells OpenAI and Anthropic to let U.S. review new models before sharing them with British testers: The White House reportedly asked OpenAI and Anthropic to withhold new models from the UK AI Safety Institute pending US review, and Anthropic has reportedly restricted Claude Mythos 5.1 access to US organisations.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedAnthropic has reportedly limited early access to its Claude Mythos 5.1 model to US organisations after the White House asked it and OpenAI to withhold new models from the UK’s AI Security Institute until a US review is complete, according to ITPro.
The report adds a concrete model and a reported policy timetable to the dispute. The US review is linked to a June executive order said to allow a 30-day federal assessment of frontier models before they are shared with international partners.
For the UK institute, the practical consequence is that pre-release testing may now depend on US approval, rather than simply on arrangements with the model developers. ITPro says British officials have pushed back and that some requests for exemptions were denied.
Sources and evidence
- OpenAI and Anthropic could withhold new AI models from UK's AI Security Institute: ITPro reports that the White House asked OpenAI and Anthropic to withhold new frontier models from the UK’s AI Security Institute pending US review, and that Anthropic has limited Claude Mythos 5.1 early access to US organisations. The report links the policy to a June executive order and alleged Commerce Department export-control directives.
Independent WittyWires Watcher; not an official account or feed.