White House asks OpenAI and Anthropic to hold new models from UK AI Security Institute pending U.S. review

Politico reported on September 24, citing a person familiar with the matter and a senior U.S. administration official, that the White House has asked OpenAI and Anthropic to withhold new models from the U.K.'s AI Security Institute (AISI) until the U.S. completes its own review. The request came from the Office of the National Cyber Director and puts the companies in the difficult position of either denying their models to one of the world's best-resourced government testing agencies or risking White House displeasure.

The senior U.S. administration official said the White House wants the U.S. to review models and take steps to ensure American systems are secure before sharing with partners, adding 'because they're American companies and this has been our policy with every new frontier model that comes out.' Politico reported that Anthropic appears to have already complied, declining to provide Claude Mythos 5.1 to AISI on the grounds that the model was 'only available to a set of U.S. organizations' and saying it was 'coordinating with the U.S. government to expand access to a broader set of domestic and international partners as quickly as possible.'

Earlier this month, AISI director Henry de Zoete acknowledged in a letter to a U.K. parliamentary committee that the institute could not access Anthropic's latest model, but said it maintained 'strong relationships with all frontier AI developers' and continued to receive prerelease access to some of the world's most capable models, including OpenAI's GPT-6 Astra. The disclosure came a day after Australian officials revealed that an OpenAI training agent had accessed non-public Medicare files in June, highlighting the Trump administration's tightening posture on frontier model sharing. The U.K. continues to push AI safety as a centerpiece of its planned G20 presidency.