The Trump administration, which framed itself as "hands off" on AI, has asked OpenAI to release GPT 5.6 to a partner group before any wider public rollout. The Office of the National Cyber Director and the White House science policy office will vet each approved user.
The White House has asked OpenAI to release its newest frontier AI model first to a small circle of partners rather than the general public, and to let federal officials approve each customer one by one. The arrangement, first reported by The Information and summarized by TechCrunch, turns the Trump administration's stated "hands-off" posture toward artificial intelligence into something more like a quiet gatekeeper.
At a staff meeting this week, OpenAI CEO Sam Altman told employees that the Trump administration would be "approving access customer by customer" during the preview period for GPT 5.6, the company's next-generation model, according to The Information's reporting cited by TechCrunch. If the limited release goes well, Altman said, OpenAI hopes to follow with a broader public release a "couple of weeks later," but only after the government signs off on the rollout.
The agencies behind the request are the Office of the National Cyber Director (ONCD), the White House office that coordinates federal cybersecurity policy, and the Office of Science and Technology Policy (OSTP), the administration's main science advisory body. OpenAI staff reportedly "worked closely" with the government on the upcoming release. The policy basis appears to be a June 2026 presidential action titled Promoting Advanced Artificial Intelligence Innovation and Security, which set the stage for interagency review of frontier models. Yahoo News also reported that the administration had asked OpenAI to stagger the launch.
The pattern is not new, but the source of it is. For comparison, Anthropic has voluntarily kept its top cyber-capability model, sometimes referred to as Claude Mythos under its Project Glasswing restricted-access program, inside a narrow partner circle instead of publishing it broadly. Project Glasswing is Anthropic's voluntary framework for limiting distribution of its most capable model to a vetted partner list, as reported in coverage of the GPT 5.6 rollout. The arrangement for OpenAI looks similar in shape. The difference is that Anthropic chose it; for OpenAI, the U.S. government requested it.
The shift is small in vocabulary and large in practice. The Trump administration originally framed itself as a release valve on Biden-era AI rules, signing executive orders that promised to clear red tape for model developers. The June 2026 action kept that deregulatory language but added a quiet control surface: voluntary pre-release submission, interagency review, and case-by-case customer approval for partner-only previews. The administration is now using that surface on the most-watched AI release of the year.
The shape of this control matters because it does not look like regulation. There is no new statute, no licensing regime, no public rulemaking. There is a request from a White House cyber office and a science policy office, a CEO telling staff the government will vet each customer, and a partner preview that could quietly become the default path to market for U.S. frontier AI. SiliconAngle's coverage frames the rollout as taking place under government vetting, though the publication's framing of a "2027 IPO" is editorial analysis rather than an OpenAI disclosure and should not be read as a confirmed date.
What to watch next is whether the partner-only preview stays a one-off for GPT 5.6 or becomes the standing template for how the administration engages other frontier labs, including Anthropic, Google DeepMind, and Meta's AI research arm. It also matters whether the preview produces any disclosed safety finding, or any public justification for the government's case-by-case approvals. Right now the vetting happens in private, the customer list is not public, and the criteria for approval have not been stated.
The administration's bet appears to be that voluntary pre-release coordination can do what formal regulation could not: slow the most capable models just enough to manage catastrophic-risk concerns, without slowing the U.S. industry's overall race against Chinese competitors. OpenAI's bet appears to be that a short, government-vetted partner preview preserves its launch timeline and keeps regulators inside the tent. Both bets are now live, and GPT 5.6 is the first frontier model being shipped through the new channel.