US Gov Approves Each GPT-5.6 Customer

OpenAI will release GPT-5.6 only to a select group of partners, with the US government approving access on a customer-by-customer basis. Commerce Secretary Howard Lutnick personally called CEO Sam Altman to warn him not to proceed without sign-offs from multiple agencies. Two offices are overseeing the process: the Office of the National Cyber Director and the Office of Science and Technology Policy, according to reporting by The Information.

Altman told OpenAI staff about the arrangement at an internal Q&A on June 26. He expects broader public access to follow a couple of weeks after the limited preview, provided nothing goes wrong in the interim. He was clear about where the company stands: "We've made clear to the U.S. 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."

How frontier AI ended up needing government sign-off

The pattern started earlier in 2026 with Anthropic. The company restricted its Mythos model through an internal effort called Project Glasswing, citing the concern that advanced AI could help attackers find and exploit software vulnerabilities faster than defenders could respond. The Trump administration followed with an executive order calling for voluntary model reviews, and most major AI labs agreed to participate. GPT-5.6 is now the second high-profile case where that framework has visibly shaped how — and when — a model reaches the public.

Altman's careful phrasing is worth reading closely. Calling this arrangement 'not our preferred long term model' is a polite but unambiguous signal that OpenAI views government oversight here as temporary — something to navigate, not accept permanently. At the same time, the company complied. That combination — compliance plus public pushback — reflects how uncomfortable the situation is for major AI labs: the government now has a say in timing, even if not in the technology itself.

What this means for developers and builders

For most people building on OpenAI's API right now, the practical effect is a delay of a few weeks before GPT-5.6 becomes widely available. That's manageable. The larger question is what happens if review requirements become routine and expand in scope. Frontier models — the most capable ones, those best suited for complex reasoning, long code, and advanced research — are the ones most likely to enter these review windows first and most consistently.

The companies best positioned to absorb a delay are large ones. A major enterprise customer can shift its roadmap by three weeks without crisis. A startup whose core product was built around the capabilities of a specific new model has a harder time. The dynamics here aren't really about any single model or any single delay — they're about whether the gap between 'model exists' and 'model available to developers' starts getting systematically longer. That would change the economics of AI development in ways that favor incumbents over challengers.