U.S. government will decide who gets to use OpenAI's newest model

OpenAI said on Friday that the U.S. government would initially approve who gets access to its newest AI model — the first time Washington has extended its vetting of AI customers beyond Anthropic. Hours later, the Commerce Department told Anthropic it could provide its latest model, Mythos 5, only to a restricted list of U.S.-based companies, according to a copy of the letter reviewed by The Washington Post.

OpenAI hands over the gatekeeping

Under the arrangement, only companies approved by the government will get access to OpenAI's new model, Sol (GPT-5.6), which the company says is its most powerful yet and shows improvements in coding and cybersecurity tasks.

OpenAI chief executive Sam Altman made clear he was not enthusiastic. "I just dont like the idea of the government picking the customers," he wrote on X, adding: "Confident we will get to a better place." In a blog post, the company said it does not believe such a government access process "should become the long-term default".

The letter to Anthropic

The Commerce Department letter, signed by Secretary Howard Lutnick, said the government had "determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model". The government reserved the right to change the list at any time, the letter said.

A person familiar with the matter said the list included around 100 companies. Anthropic said it had been told it could redeploy Mythos 5 to a "small group of cyber defenders and infrastructure providers" and was working to restore their access.

The move came two weeks after the administration barred Anthropic from giving non-U.S. citizens — including its own employees — access to Mythos 5 and Fable 5, prompting the company to withdraw the models.

Why the shift matters

The measures mark a rapid change in U.S. AI policy. President Donald Trump returned to office promising a hands-off approach and criticised the Biden administration's efforts to set safety standards; after systems capable of finding security vulnerabilities in software alarmed officials at home and abroad, the White House changed position.

"In a matter of weeks, U.S. federal AI policy has gone from implausibly libertarian to increasingly draconian and opaque," wrote Dean Ball, a former Trump AI adviser who has said he will join OpenAI next month to work on policy.

The effects are being felt beyond Washington. A British MP said the U.K.'s AI Security Institute had access to OpenAI's new GPT-5.6 model and, according to a person familiar with the arrangement, is the only non-U.S. entity with access. At this month's Group of Seven summit in France, the leaders of France and Canada raised concerns about over-reliance on U.S. AI providers.