The White House's Office of the National Cyber Director and the Office of Science and Technology Policy have asked OpenAI to restrict the initial release of its next model, GPT-5.6, to a small set of government-approved partners before any broader distribution, according to a person familiar with the matter.
Why this matters
This request is the first known instance of the U.S. government asking an American AI company to curb a model's launch prior to release. It reflects efforts by the administration to develop a framework for testing and evaluating the security of new models.
The immediate news
The Information reported earlier on Thursday that OpenAI CEO Sam Altman informed employees in a memo about plans for a limited rollout. According to The Information, Altman wrote that the company had made clear to the U.S. government this was not its preferred long-term approach and that it would work with the government and industry to establish a more sustainable process for future releases.
A source told Axios that OpenAI had been proactively coordinating with the administration on the model release even before Anthropic's access to its frontier models, Fable 5 and Mythos 5, was revoked following a rare directive from the Commerce Department.
Government preview and discussions
The source said the White House had been briefed on GPT-5.6's capabilities and had an opportunity to preview what the model can do. Axios learned that Sam Altman discussed GPT-5.6 with Commerce Secretary Howard Lutnick on Wednesday; Lutnick wanted assurance that all relevant parts of government would test and approve the model.
The source characterized the government's intervention as precautionary because GPT-5.6 has "Mythos-like" capabilities, not as an indication the administration is suddenly taking a heavier regulatory hand. "This is what's happening with models of that caliber," the source said, adding that the models are powerful enough that the government wants to confirm companies have adequate safeguards in place.
Related policy context
Earlier this month, President Trump signed an AI security executive order directing several agencies to set up a voluntary testing protocol for AI companies before they release a new model. Political disputes over how strict and mandatory that program should be delayed implementation for weeks.
The broader picture
AI labs face pressure to release upgraded models quickly to stay competitive — including against increasingly capable Chinese open-source models — while addressing rising security concerns. Security officials and corporate leaders worry about the consequences if bad actors, such as nation-state spies, cybercriminals, or malicious insiders, obtain access to highly capable models.
What to watch next
According to The Information, Altman wrote in the memo that he hopes to make GPT-5.6 available more broadly "a couple of weeks later." The outcome of government testing and approvals will be critical to the model's wider release schedule.
