The Trump administration has asked OpenAI to delay and stagger the release of its next frontier model, GPT-5.6, limiting initial access to a small circle of government-vetted partners that federal officials will approve one customer at a time. The request, reported on June 25, 2026, marks the first time the U.S. government has preemptively moved to restrict the launch of an American AI model before it reaches the market.
According to a staff memo from CEO Sam Altman, the government will be "approving access customer by customer" during a preview period, with a broader public release that OpenAI hopes will follow "a couple of weeks later." The discussions involved officials from the Office of the National Cyber Director and the Office of Science and Technology Policy, with Commerce Secretary Howard Lutnick also part of the conversation.
Key Highlights
- The White House asked OpenAI to stagger GPT-5.6's release rather than launch it broadly, citing cybersecurity and national security concerns.
- Initial access is limited to a small group of government-approved partners, vetted individually on a customer-by-customer basis.
- It is the first time Washington has stepped in to gate an American AI model before public release.
- The move follows a June 2, 2026 executive order establishing a voluntary framework that gives federal cybersecurity teams up to 30 days to assess frontier models.
- Altman told staff this is "not our preferred long term model" for how releases should work.
Details
The intervention is rooted in a broader policy shift. On June 2, 2026, President Trump signed an executive order, "Promoting Advanced Artificial Intelligence Innovation and Security," directing federal agencies to build a framework for the secure deployment of frontier AI. Under that framework, developers can voluntarily give the government early access to new models for up to 30 days so cybersecurity teams can evaluate them before they go live for partners and the public.
GPT-5.6 is the first major model to run through this process in practice. Reports indicate the initial preview will be restricted to roughly 20 trusted, government-vetted partners, with officials from the National Cyber Director's office and the Office of Science and Technology Policy approving enterprise customers case by case.
The official rationale centers on capability. Officials are concerned that the newest generation of models can identify and exploit software vulnerabilities at unprecedented speed, raising the risk that bad actors or foreign adversaries could weaponize them against critical infrastructure if released without adequate vetting.
OpenAI's Response
Altman framed cooperation as the fastest path to eventually getting GPT-5.6 into broad availability, while making clear the company sees the arrangement as a stopgap, not a template. "We've made clear to the US government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases," he reportedly told employees.
Neither the White House nor the National Cyber Director's office responded to requests for comment.
Impact
For OpenAI, the staggered rollout adds a regulatory checkpoint to a launch cadence that has, until now, been governed by the company's own timelines. For the wider industry, it signals that frontier AI is increasingly being treated as strategic national infrastructure — closer to semiconductors, defense technology, or nuclear research than to consumer software.
The decision also reshapes how enterprises plan around new models. If access depends on government approval granted customer by customer, procurement timelines, pilot programs, and product roadmaps that assume day-one availability now carry new uncertainty.
Background
The GPT-5.6 request does not stand alone. Earlier in June 2026, the administration moved against rival Anthropic, with the company restricting access to its most capable Fable 5 and Mythos models after researchers demonstrated ways to bypass safety guardrails. OpenAI's staggered approach echoes Anthropic's earlier limited rollout of Mythos through a restricted-partner program tied to cybersecurity concerns.
Together, the two episodes within a single month suggest a durable change in posture: the era of frontier models shipping broadly on launch day, on the developer's schedule alone, is giving way to government-mediated release windows.
What's Next
OpenAI hopes the limited preview proceeds smoothly enough to justify a broader release within weeks. Beyond GPT-5.6, the larger question is whether the voluntary 30-day review framework hardens into a standing requirement — and how it will interact with the open-weight models that face no such gate.
For the MENA region, where organizations increasingly weigh AI access against data-protection regimes such as Tunisia's INPDP and Saudi Arabia's PDPL, the development underscores a growing reality: access to the most capable models is becoming a question of geography, governance, and approval, not just price and API keys. Enterprises building on frontier models will need to factor regulatory availability — and credible self-hostable alternatives — into their planning.
Source: TechCrunch, Engadget, Axios, The Decoder