The US government requests that OpenAI delay the release of its upcoming model.

      Sam Altman informed employees that Washington wants GPT-5.6 to be released first to a select group of trusted partners, with access granted on a customer-by-customer basis.

      For years, the discussion surrounding the slowdown of powerful AI models has been the responsibility of company safety teams and external critics. Now, however, it includes a government request.

      The Trump administration has asked OpenAI to stagger the release of an upcoming model, marking the first instance where the US government has proactively requested an American AI company to limit a launch ahead of its occurrence.

      This direction came from the top, as OpenAI's CEO Sam Altman communicated to staff on Wednesday that the government had requested the company initially launch the model to a short list of trusted partners before a broader rollout.

      During this preview period, Altman indicated that the government would be “approving access customer by customer.”

      The request didn’t stem from a single department; it was based on discussions with two government offices, the Office of the National Cyber Director and the Office of Science and Technology Policy, which emphasize cybersecurity concerns over issues of competition or content.

      The primary concern articulated is what a sufficiently advanced model could do if misused, and the staggered rollout is intended to minimize that risk during the early phase.

      The timing of this request reflects a wider change. It follows closely after rival Anthropic had its most advanced offerings withdrawn from the market under a government directive, suggesting that Washington is now actively influencing the release schedules of leading laboratories rather than merely reacting to them after the fact.

      The proposed process is significant in itself. A customer-by-customer approval procedure during the preview phase would, if performed as described, allow a government agency direct involvement in determining who receives early access to a cutting-edge model.

      This approach mirrors the controlled rollout OpenAI implemented for GPT-5.4-Cyber, which was released to vetted security teams under a Trusted Access program.

      Such a method represents a significantly more proactive stance compared to the voluntary commitments and post-facto evaluations that have characterized US AI policy up to now, shifting at least temporarily the control over a release from the company to the state.

      For OpenAI, this arrangement presents multiple consequences. The staggered rollout delays the company’s capacity to present its newest model to paying customers and developers, coming only months after it introduced GPT-5.5 to the enterprise market, which incurs commercial costs in a rapidly moving market.

      Additionally, it provides a layer of political protection: a model launched with the government’s clear involvement makes it more difficult to hold the company accountable if issues arise.

      How OpenAI balances these factors will become clearer once the preview period and its subsequent phases begin.

      Much of the information still relies on Altman’s comments to staff and reports from various sources rather than an official statement from the government, and OpenAI has not released the terms of the arrangement.

      Details such as the model name, the customer approval process, and the involved agencies are derived from these accounts.

      What this situation establishes, if it proves accurate, is a new approach: a US administration treating the release of a cutting-edge model as something that should be gated, with a leading lab conforming to this request.

      The following question is whether this will set a standard for future releases.