Sam Altman told staff Washington wants GPT-5.6 released first to a short list of trusted partners, with access approved customer by customer.
For years the debate over slowing down powerful AI models was a matter for company safety teams and outside critics. Now it has a government request attached.
The Trump administration has asked OpenAI to stagger the release of an upcoming model the first time the US government has preemptively asked an American AI company to restrict a launch before it happens.
The instruction reached employees from the top. OpenAI chief executive Sam Altman told staff on Wednesday that the government had requested the company initially release the model to a short list of trusted partners before pushing it out more widely.
The government, Altman told staff, would be “approving access customer by customer during this preview period.”
The request did not come from a single office. According to the reporting, it emerged from conversations with two government bodies, the Office of the National Cyber Director and the Office of Science and Technology Policy, which frames the concern as one of cybersecurity rather than competition or content.
The worry, as described, is what a sufficiently capable model could do in the wrong hands, and the staggered rollout is meant to limit that exposure during an initial window.
The timing places the request inside a wider shift. It comes roughly two weeks after rival Anthropic saw its most capable offerings pulled from the market under a government directive, which suggests Washington is now actively shaping the release schedules of the leading labs rather than reacting to them after the fact.
The mechanism described is notable in its own right. A customer-by-customer approval process during a preview period would, if it operates as reported, give a government agency a direct hand in deciding who gets early access to a frontier model.
It echoes the gated rollout OpenAI used for GPT-5.4-Cyber, released to vetted security teams under a Trusted Access programme.
That is a markedly more hands-on posture than the voluntary commitments and after-the-fact evaluations that have characterised US AI policy to date, and it shifts the locus of control over a release, at least temporarily, from the company to the state.
For OpenAI, the arrangement cuts in more than one direction. A staggered rollout slows the company’s ability to put its newest model in front of paying customers and developers, only months after it launched GPT-5.5 into the enterprise market, which carries a commercial cost in a market where rivals move quickly.
It also offers a measure of political cover: a model released with the government’s explicit involvement is harder to blame the company for if something goes wrong.
How OpenAI weighs those against each other will become clearer once the preview period, and whatever follows it, is underway.
Much of the detail still rests on Altman’s account to staff and on reporting from sources rather than an official government statement, and OpenAI has not published the terms of the arrangement.
The model name, the customer-by-customer approval mechanism, and the agencies involved come from those accounts.
What the episode establishes, if it holds, is a new posture: a US administration treating a frontier model’s release as something to be gated, and a leading lab agreeing to the gate.
The next question is whether this becomes the template for every release that follows.
