The Trump administration has asked OpenAI to stagger the release of its next model, GPT-5.6, over cybersecurity concerns, according to The Information. Instead of a broad public launch, OpenAI will make the model available in a limited preview to a small group of partners, with the government approving access on a customer-by-customer basis during the preview period.

A delay to GPT-5.6 had been circulating as a rumor in recent days. The Information’s report is the first sourced account of why, and of the customer-by-customer arrangement OpenAI has agreed to.

The request came out of conversations with two government bodies, the Office of the National Cyber Director and the Office of Science and Technology Policy. CEO Sam Altman told staff in a Q&A that the government would sign off on each customer individually during the preview, an unusual arrangement for a commercial software product. The reporting cites security concerns as the rationale but does not specify the vulnerabilities involved.

The move is the clearest application yet of the framework the administration put in place earlier in June. The executive order on frontier AI created a process for the government to review “covered frontier models” before release and to designate trusted partners for early access, with a classified benchmark of a model’s cyber capabilities run through the National Security Agency. Staggering GPT-5.6 and gating who may use it during a preview is that process applied to a leading US lab’s flagship model.

It also follows a more abrupt intervention. The same administration ordered Anthropic’s most capable models switched off days after launch on national-security grounds, and is separately negotiating a trusted-partner scheme to govern allied access to American models. A staggered GPT-5.6 extends the same logic from Anthropic to OpenAI: the government is increasingly deciding not just whether a frontier model can ship, but who can use it and when.

OpenAI and the two agencies have not publicly confirmed the arrangement, and the specific concerns behind it have not been disclosed. What is described is voluntary, but the practical effect is that a flagship commercial model would reach the public on the government’s schedule rather than the developer’s.

Sources: The Information, Reuters via Yahoo, Quartz

–
By the Control Plane Editorial Team