White House asks OpenAI and Anthropic to wait for a US review
The White House asked OpenAI and Anthropic to hold new models from British testers until a United States review, CNA reported via Politico and City A.M. confirmed from the Office of the National Cyber Director. City A.M. says Anthropic kept Claude Mythos 5.1 initially inside US organisations, while the AISI director said the institute had already tested OpenAI's GPT-6 Astra. CNA added that the companies and the White House did not immediately comment to Reuters.
Artificial Intelligence··Evening
Washington asks for the first look
The White House has asked OpenAI and Anthropic to hold new models back from British testers until a United States review, CNA reported on 25 September, citing Politico. City A.M. describes the same request: the companies were asked to give Washington the first look before sharing models with Britain's AI Security Institute, the AISI. City A.M. says the request came from the Office of the National Cyber Director. A senior US administration official told that paper the aim was to be sure American systems were secure before models reached allies, and that this had been the policy with every new frontier model because the firms are American companies.[1], [2]
Anthropic limits the newest model
City A.M. reports that Anthropic appears to have followed the request. The company did not provide its latest Claude Mythos 5.1 model to the AISI and said access was initially limited to US organisations. Anthropic added that it was coordinating with the US government to widen access to a broader set of domestic and international partners as quickly as possible. CNA's account stops shorter. It says the stated reason for the hold is to check that US systems are secure before models are shared with partners, and that the White House, Anthropic and OpenAI did not immediately answer a request for comment from Reuters.[2], [1]
The institute already tested an earlier OpenAI model
City A.M. quotes AISI director Henry de Zoete, who acknowledged this month that the watchdog had not received Anthropic's latest model, while saying the institute kept strong relationships with frontier developers. He added that the institute tested OpenAI's GPT-6 Astra before its release. The two reports meet on the hold itself. CNA, via Politico, and City A.M. both say Washington asked OpenAI and Anthropic to wait for a US review before British testers saw new models. City A.M. alone names the cyber director's office, Mythos 5.1 and the Astra test.[2], [1]
