For most of the AI era, the question "who gets to use this model?" has been answered by the company that built it. This week, for one of the most capable models yet, a different answer arrived: the US government will decide — one customer at a time.

According to a report from The Information, OpenAI has agreed to stagger the release of GPT-5.6 after the Trump administration formally asked it to limit who gets early access. During the preview period, access will reportedly be approved by the government customer by customer, with the model going initially to only a small group of vetted partners — around 20, according to The Information. The request, the report says, came from the Office of the National Cyber Director and the Office of Science and Technology Policy, citing the model's advanced capabilities and national-security implications.

The detail that makes this notable is in an internal memo from OpenAI CEO Sam Altman, quoted by The Information. Altman reportedly told staff the government would be "approving access customer by customer during this preview period" — and, tellingly, added: "We've made clear to the U.S. government that this is not our preferred long-term model, and will work with them and others in industry to achieve a more sustainable approach for future releases." A frontier lab complying with federal access controls while openly signalling it doesn't want them to become the norm.

Why this is a bigger deal than it sounds

It's worth being precise about what's new here, because it's easy to file under "more AI regulation" and move on. This is not an export control — the kind of restriction, like the one placed on Anthropic's models earlier this month, that limits who abroad can buy a model. This is the government inserting itself into who gets access to a frontier model at all, including domestically, before it ships. The gatekeeping has moved from the border to the release itself.

Set it against the rest of the week and a pattern is hard to miss. The Five Eyes intelligence alliance warned that AI-driven cyberattacks are "months, not years" away. Anthropic accused a Chinese lab of siphoning its models' capabilities. And now the US government is pre-clearing who gets early access to OpenAI's newest model on national-security grounds. Frontier AI is increasingly being treated less like a software product and more like a controlled technology — closer to encryption or advanced chips than to a consumer app.

The caveats worth keeping

A few things to hold onto before drawing big conclusions. This is, by the reporting, a preview-period arrangement, not a permanent regime — and Altman's memo frames it as something OpenAI expects to renegotiate. The specifics (the customer-by-customer approval, the ~20 partners, the two federal offices involved) come via The Information's reporting and an internal memo, not a published government order, so treat the precise mechanics as well-sourced reporting rather than official text. And "staggered access for a preview" is, on its own, fairly common for powerful new models — what's striking is who is now holding the gate.

The takeaway

The headline isn't that GPT-5.6 is delayed; staggered rollouts are routine. It's that, for the first time in a way this visible, a branch of the US government is deciding which companies get early access to a leading commercial AI model — and a leading AI company is going along with it, if reluctantly. Whether this was a one-off born of a specific security worry or the first draft of how frontier models get released from now on is the question that matters. Altman's own words suggest even OpenAI isn't sure — and would rather it were the former.