The US government requests that OpenAI delays the launch of its upcoming model.

      Sam Altman informed employees that the government wants OpenAI to initially release GPT-5.6 to a select group of trusted partners, with access granted on a customer-by-customer basis. For years, the debate surrounding the slowing down of powerful AI models was primarily a concern for company safety teams and external critics. Now, it involves a request from the government.

      The Trump administration has asked OpenAI to stagger the rollout of its forthcoming model, marking the first instance where the US government has proactively requested an American AI firm to limit a launch prior to its occurrence. This directive was communicated to staff by OpenAI's CEO, Sam Altman, who stated that the government requested the initial release to be restricted to a limited number of trusted partners before general availability.

      Altman explained that during this preview phase, the government would be responsible for "approving access customer by customer." The request originated from discussions with two government agencies, the Office of the National Cyber Director and the Office of Science and Technology Policy, framing the concern in terms of cybersecurity rather than competition or content.

      The primary issue at hand is the potential risks a sufficiently capable model could pose if misused, prompting the staggered rollout to minimize exposure during the initial phase. This request aligns with a broader trend, occurring about two weeks after rival company Anthropic had its advanced offerings withdrawn from the market due to a government directive, indicating that Washington is now actively influencing the release schedules of top labs instead of merely responding to them.

      The proposed approval process, where access would be granted on a customer-by-customer basis during the preview phase, would provide a government agency with direct input on who gets early access to a cutting-edge model. This approach is reminiscent of the gated rollout OpenAI employed for GPT-5.4-Cyber, which was released to selected security teams under a Trusted Access program.

      Such a stance represents a more direct intervention compared to the voluntary commitments and retrospective assessments that have characterized US AI policy thus far, shifting control over releases, at least temporarily, from the company to the government.

      For OpenAI, this situation has mixed implications. The staggered rollout may hinder the company’s ability to showcase its latest model to paying customers and developers, particularly just months after launching GPT-5.5 in the enterprise market, which could have financial repercussions in a competitive space. Conversely, it also provides some political protection: a model released with the government’s explicit involvement mitigates the company's liability if issues arise.

      How OpenAI balances these factors will become evident once the preview period and any subsequent phases commence. Much of the information is still based on Altman's remarks to staff and reports from sources, rather than an official government release, and OpenAI has not disclosed the specifics of the arrangement. The model name, the customer-by-customer approval process, and the agencies involved are drawn from these accounts.

      This situation establishes, if it holds, a new approach: a US administration treating the release of a leading model as something to be gated, with a prominent lab consenting to such restrictions. The next question is whether this will set a precedent for subsequent releases.