The Government Made OpenAI Wait. First Time for Everything.

The Trump administration asked OpenAI to delay the release of GPT-5.6 last week. OpenAI said yes.

That's not how this usually goes.

The standard AI launch playbook goes like this: company announces model, company releases model, regulators eventually notice, Congress holds hearings featuring senators who don't know what a token is, nothing substantive happens. We've run this loop about fifteen times now. You know the script.

What happened with GPT-5.6 is different. The White House's Office of the National Cyber Director and Office of Science and Technology Policy reached out ahead of the release. They wanted OpenAI to limit the rollout to a small set of government-approved partners while federal agencies reviewed what the model can do. Commerce Secretary Howard Lutnick was apparently in direct conversations with Sam Altman.

Altman complied. He sent a memo to employees saying GPT-5.6 would go to enterprise partners first, with a broader release a 'couple of weeks' later. He also made clear this is not a precedent OpenAI wants to set.

That last part is the interesting sentence.

What Altman Actually Said

The internal memo, according to reporting from TechCrunch and Axios, had Altman telling employees that this arrangement 'is not our preferred long-term model.' He framed it as a one-time accommodation while OpenAI works with the government on something more sustainable.

Translation: we'll do this once. We're not building a system where every new model needs White House sign-off.

That's not unreasonable. OpenAI operates in an intensely competitive market. Anthropic didn't pause Claude. Google didn't ask permission before launching Gemini 2.5 Pro with Deep Think. Frontier model releases are accelerating, not slowing down. If OpenAI has to pause for government review every time while competitors don't, that's a structural disadvantage, not a safety measure.

So Altman cooperated, drew a line around it being temporary, and moved on. That's the right call. But the fact that it happened at all is worth examining.

The Enormous Irony Here

Let me walk you through the timeline, because it's good.

In early 2025, the Trump administration revoked Biden's AI executive order. That order, among other things, required companies developing powerful models to share safety test results with the government before release. It set up evaluation frameworks. It built in mandatory disclosure mechanisms. The Trump administration called it regulatory overreach and killed it.

So now we're in a world where there's no formal pre-release review requirement.

And the Trump administration just informally asked OpenAI to do a pre-release review.

They tore down the stop sign and are now standing in the intersection asking cars to stop.

I want to be precise here: this isn't necessarily hypocritical in the way it might seem. The Biden order was a standing rule applying to all covered models. What the Trump administration did was a targeted ask, applying to one model, managed through direct relationship rather than regulation. That's a different governance philosophy. You can think regulation is bad and still think individual conversations with companies are appropriate.

But that philosophy has a name. We're governing AI by relationship, not by rule. The outcome of any particular conversation depends on who's in the room, how much leverage they have, and whether the company feels like cooperating that day.

In this case: OpenAI cooperated. GPT-5.6 gets a few weeks of limited release. What happens when the next model is more capable, the timeline is tighter, and the company decides the commercial cost is too high? There's no rule to appeal to. There's just a phone call.

What the Government Is Actually Worried About

The stated concern is national security. New frontier models can meaningfully advance capabilities that matter for defense: sophisticated code generation, persuasion at scale, vulnerability identification. If GPT-5.6 significantly improves on any of these, you'd want to know that before it's in everyone's pocket.

That's legitimate. European officials recently warned that pro-Russian groups are actively seeding AI chatbots with propaganda to influence American elections. Chatbots are not neutral tools. What a model can do at scale has real consequences. A government wanting to understand a model before broad release isn't crazy.

The question is process. An informal phone call is not a process. It worked this time because OpenAI felt enough goodwill toward the administration to cooperate. It might not work with the next lab. Or the next model. Or if the relationship cools.

Meanwhile, the formal evaluation infrastructure that was being built is gone. We traded durable rules for personal relationships. In tech policy, that tends to age poorly.

What Happens Next

Altman said 'a couple of weeks,' so GPT-5.6 probably has a broad release in mid-July. This becomes a footnote. A historically significant one: the first time a U.S. administration preemptively asked an AI lab to hold a model. But still a footnote, unless something changes.

Nobody's pushing for a formal review process. The government hasn't said it wants one. OpenAI hasn't volunteered to build one. Everyone is treating this as a one-off exception.

Which means the next exception is just a phone call away.

The government made OpenAI wait. OpenAI said fine, for now. And that's apparently where we are: the most powerful AI company in the world, building systems capable of affecting elections and infrastructure, operating under a governance model that amounts to 'we'll call you if we're worried about this one.'

Good luck to all of us.

Want one weekly email covering AI without the noise? Subscribe to About.chat Weekly.