GPT-5.6, the White House slows down OpenAI: staggered release and one-by-one approved access

The Trump administration requested OpenAI to stagger the release of GPT-5.6 for security reasons, and the company agreed. The new model will not be available to the general public right away: in the initial phase, it will be available only to a select group of enterprise customers and partners, during which time the federal government will approve access on a customer-by-customer basis. This request was revealed by The Information.

The federal agencies that made the request are the Office of the National Cyber Director and the Office of Science and Technology Policy. Reportedly, OpenAI staff have been working closely with the government in anticipation of the model's imminent release. If the limited phase goes smoothly, the company aims for a broader opening in “a couple of weeks.”

The decision communicated by Altman

It was Sam Altman himself who communicated the decision to employees during a company Q&A session. The head of OpenAI emphasized that this is a temporary solution: "We have made it clear to the U.S. government that this is not our preferred mode in the long term, and we will work with them and other industry players to arrive at a more sustainable approach for future releases." The wording implies that the company accepts the constraint for this launch but does not intend to make it a standard practice.

Trump had already signed an executive order encouraging companies in the sector to voluntarily submit new models to the government for testing and evaluation before public distribution. The GPT-5.6 issue illustrates that principle applied to a concrete case, with a degree of control over access that goes beyond mere preliminary assessment.

The comparison with the Anthropic case

The treatment reserved for OpenAI appears to be softer than that faced by its direct competitor, Anthropic, which received a federal directive that blocked access to its leading models for anyone who is not a U.S. citizen, including its own employees. In comparison, the request for a staggered release with one-by-one approved access is a less drastic limitation, keeping the operational model in the hands of real customers.

The two episodes seem to signal a tightening of federal oversight over frontier models, though articulated in varying intensities case by case. For now, it remains to be seen how long the restricted phase of GPT-5.6 will last: OpenAI's indicated timeline speaks of a few weeks, but the green light for a broader opening depends on the outcome of the preview and the approval of the involved agencies.