White House orders OpenAI to stagger GPT-5.6 launch and approve users one by one

OpenAI had been readying the release of GPT-5.6, its next frontier language model. A general launch was on the cards. That changed this week when chief executive Sam Altman gathered staff on Wednesday and delivered a message from Washington: the government wanted the rollout slowed and the initial pool of users hand-picked by federal officials.

The White House, acting through the Office of the National Cyber Director and the Office of Science and Technology Policy, asked OpenAI to put GPT-5.6 into a limited preview for a short list of trusted partners. During that preview, access would be granted only after the administration approved each customer individually. Altman described the arrangement as not the company's preferred path but said he was hopeful a smooth preview could lead to a broader release a couple of weeks later.

The Anthropic precedent

The instruction did not arrive in a vacuum. Two weeks earlier, rival lab Anthropic had pulled both of its new models – Fable 5 and Mythos 5 – from the market entirely after a government directive. The administration told Anthropic to block any foreign nationals from using the tools, citing national security concerns. Unable to verify customer nationalities reliably, Anthropic disabled global access, a far more drastic restriction than OpenAI now faces.

Government's case-by-case control

The customer-by-customer vetting model is new in its directness. Until now, the US government's role in AI releases had been limited to voluntary safety testing and post-launch review. Under the arrangement reported for GPT-5.6, a federal agency effectively becomes a gatekeeper for who gets early access to a frontier model. The concern is that a model capable of finding software vulnerabilities could, in the wrong hands, be weaponised for cyberattacks. OpenAI itself used a gated release for its earlier GPT-5.4-Cyber model, but that was a company initiative, not a government requirement.

We've been working closely with the government on this release. They will be approving access customer by customer during the preview period.

Altman's reluctant optimism

Altman told staff the company had made its discomfort clear. "We've let the government know this isn't how we'd prefer to launch a model," he said, according to an internal memo seen by The Information. Still, he struck an upbeat note: if the controlled preview goes well, a general launch could follow a matter of weeks later. The timeline now depends as much on Washington's assessment as on OpenAI's engineering roadmap.

If the limited release goes well, we hope to do a broader launch a couple of weeks later.

A wider regulatory shift

The move fits into a pattern. In early June, President Donald Trump signed an executive order asking AI firms to submit their most capable models for federal testing ahead of public release. That order was voluntary on paper, but the interventions at Anthropic and now OpenAI suggest a much more assertive posture. Agencies such as the Treasury and Commerce departments have also been involved in the GPT-5.6 discussions, according to the Financial Times, signalling that multiple arms of government are coordinating on AI access.