OpenAI Staggers GPT-5.6 Behind 'Customer-by-Customer' Approval at the US Government's Request

OpenAI limited access to its next model, GPT-5.6 (codenamed Sol), so that it can only be used after customer-by-customer approval during a preview phase, The Decoder and The Next Web reported on June 26, 2026. According to the reports, Sam Altman told staff in an internal message that the US government asked OpenAI to narrow initial access, and OpenAI said it would roll out by approving customers one at a time during the preview window. The agencies cited include the Office of the National Cyber Director and the Office of Science and Technology Policy, with cybersecurity cited as the rationale — though OpenAI has not published the terms, and there is no official government statement, so much of this rests on Altman's account to staff and reporting by The Information. This article separates the confirmed reporting from what remains unverified.

An Unusual Preview Paced by a Government Request

OpenAI previewed GPT-5.6 (codenamed Sol) and chose a staggered rollout that narrows initial access, The Decoder and The Next Web reported. According to the reports, Sam Altman told staff in an internal message that the US government asked OpenAI to slow the release and limit access. OpenAI plans to open the model first to a short list of trusted partners and to defer broader availability by a couple of weeks. It is an unusual case in which a flagship release is paced by a government request rather than the company's own schedule.

Why the Control Targets Access, Not the Model

The core of the US government's request is customer-by-customer approval during the preview period. The Decoder reported that GPT-5.6 access requires sign-off customer by customer during the preview, with OpenAI hoping for broader release a couple of weeks later. The Next Web summarized that initial access is restricted to a short list of trusted partners, with government approval required for each customer. Rather than blocking the model itself, the structure controls who gets to use it first.

Who Intervened and on What Grounds

The agencies cited in the request include the Office of the National Cyber Director and the Office of Science and Technology Policy. The Decoder reported that Commerce Secretary Howard Lutnick called Altman to warn against proceeding without sign-off from additional agencies. The rationale for the staggered rollout is cybersecurity, framed as reducing the risk of what a sufficiently capable model could do in the wrong hands. Reporting traces the request to a Trump administration executive order calling for voluntary AI model safety reviews.

Why This Structure Carries Weight

What stands out is the direction of control. Government involvement has usually looked like after-the-fact regulation or voluntary pledges, but here the decision of whom to open the model to becomes something approved before release. In effect the access gate is operated alongside the government, which suggests that frontier-model deployment can be folded into a national-security procedure rather than a purely commercial call. The cybersecurity framing is worth watching too. Once a model's risk is defined as potential for misuse, the more capable the model, the greater the incentive to delay its release. It reads as a signal that the balance between openness and control is tilting, slowly, toward control.

How Korean Practitioners Should Read This

For teams building products on frontier models via API, the implications are clear. If a structure of trusted-partner lists and customer-by-customer approval hardens for initial access, being first to use the latest flagship becomes a question of approval queue position rather than engineering skill. Where a customer outside the US sits in that queue cannot be known from this reporting alone. But with no confirmed terms at this point, pinning a roadmap to the latest model's release date is risky. Benchmarks, pricing, and a model card are all still unconfirmed, so it is safer not to build performance expectations on nothing.

What Has Not Been Confirmed

OpenAI has not published the specific terms of the arrangement in any official document, and that gap should be stated plainly. The Next Web noted that much of the detail rests on Altman's account to staff and on source-based reporting, with no official government statement. Performance figures for GPT-5.6, such as benchmarks, pricing, and a model card, are not confirmed in these reports. As for Altman's words on the request, his message reportedly stated, "We've made clear to the U.S. government that this is not our preferred long term model."

Sources: The Decoder · The Next Web · OpenAI