White House asks OpenAI and Anthropic for US review of models before sharing with British AI Safety Institute
According to a report, the White House asked OpenAI and Anthropic to have new AI models reviewed by a US agency first, before providing them to the British AI Safety Institute. Anthropic has already complied and made Claude Mythos 5.1 available only to US organizations.
According to information from the Office of the National Cyber Director, the White House asked OpenAI and Anthropic to delay giving the British AI Safety Institute (AISI) access to new models until US agencies have reviewed them. The companies thus face a choice between restricting access for AISI and risking the White House's displeasure. According to the source, Anthropic has already complied with the request and made Claude Mythos 5.1 available only to organizations in the US.
According to the source, AISI is among the best-equipped government institutions in the world for testing AI and had earlier access to the models of leading labs. The agency was the first to report cases of autonomous deception by AI agents in real-world operation. AISI director Henry de Zoete stated in a letter to the UK Parliament that the agency still has access to leading models and tested the GPT-6 Astra model before its release.
British Prime Minister Andy Burnham downplayed the dispute at the UN General Assembly and called for "shared global principles and standards." Meanwhile, according to the source, the US agency responsible for the review, the Center for AI Standards and Innovation (CAISI), has no permanent director and has only a few dozen employees.
See the source article for details.
Why it matters
If the described procedure is confirmed and extended to other models, organizations outside the US may face delayed or limited access to the latest AI models, since their release would depend on approval by a US agency. At the same time, it suggests a possible fragmentation of international cooperation on AI safety testing between the US and the United Kingdom.
Relevant practical impact
What this means
For a business
Companies outside the US may encounter limited availability of the latest models (in the case described, Claude Mythos 5.1 only for US organizations) if a similar approval process is applied to other versions or providers as well.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.