Sam Altman posted a message tonight that looks reasonable on the surface. GPT-5.6 Sol is launching, he said — smart, efficient, a significant step forward. Also Terra, with 5.5-level performance at half the price. Good news, right?
Then the other shoe: the US government asked OpenAI to launch in a limited preview instead of open access. Customer by customer approval. Government gatekeeping, plain and simple.
Altman’s response? “I think it is quite reasonable to roll out models in this way.”
Bullshit.
The Spin Machine Is Running
Let me translate what Altman actually wrote. He said the government-mandated limited launch “fits with our long-held strategy of iterative deployment.”
Iterative deployment. That is the same phrase OpenAI has used for years to describe launching products fast, getting them in users’ hands, and fixing problems in the wild. In practice, it always meant release when it is ready for market, not when it is safe.
Now the government is doing the iterating, approving customers one by one. So iterative deployment means something totally different. It means: we release what the government lets us release, to whoever the government approves.
Altman wants you to believe this is the natural evolution of their strategy. It is not. Rather, it is a brand retrofit for a reality he cannot escape.
The post does what all good PR does. It makes a loss of control sound like a strategic choice. In addition, it reframes the government’s hand as a shared mission. “We want to be a reliable, dependable partner that works with all stakeholders.” That sentence could have been written by a White House press secretary.
Sam Altman Was Fired for Safety Posturing
This is the part that gets me.
Sam Altman was ousted from OpenAI in November 2023 by a board that included the company’s own chief scientist, Ilya Sutskever. The board said Altman was not “consistently candid.” The subtext, widely reported: they did not trust him to make responsible decisions about when and how to release increasingly powerful AI.
Still, he came back five days later because 700 out of 770 employees threatened to walk. Not because he proved he could be trusted. Because he proved he had more power than the people who were supposed to oversee him.
Since then, Altman has shipped GPT-4, GPT-4o, GPT-4.1, GPT-5, GPT-5.2, GPT-5.5, and now GPT-5.6. Seven major model launches in roughly two years. “Iterative deployment” in practice has meant: push the frontier forward as fast as engineering allows, let the market absorb the risk, and let the safety researchers file their reports after the fact.
Now the same government that Altman hoped would leave him alone is asking him to slow down. But he is not fighting it. Instead, he is praising it.
The June 2 Executive Order Was Not Voluntary
First, let us not pretend this is a surprise. The June 2 executive order on “secure deployment of frontier AI models” laid out a framework described as voluntary. Developers would give the government early access up to 30 days before release. Voluntary.
Two weeks later, the government effectively killed Anthropic’s Fable 5. Now OpenAI is doing customer-by-customer government approval, as TRT reported yesterday.
The pattern is unmistakable. In fact, the line between voluntary and mandatory vanished the first time the government said no and the company complied. Every subsequent request is a demand dressed in collaborative language.
Altman knows this. That is why his post is so carefully calibrated. He is not going to be the CEO who fights the government and loses. He has watched what happened to Anthropic. Furthermore, he has read the same executive order everyone else has. He knows his IPO ambitions, his non-profit-to-for-profit restructuring, and his entire corporate strategy depend on staying in the government’s good graces.
So he smiles and calls it reasonable.
What He Did Not Say
The most revealing parts of Altman’s post are the things he carefully avoided.
First, he did not say when GPT-5.6 would be generally available. “As fast as we can” is not a timeline.
Second, he did not explain how customer-by-customer government approval squares with OpenAI’s mission of “benefiting all of humanity.” If the government decides a customer is too risky, does that customer get told why? As a result, there is no appeals process yet.
Third, he did not acknowledge that his own company’s safety track record — the boardroom coup, the non-disparagement agreements, the lawsuits, the models rated High in cybersecurity and bioweapons risk categories — may have contributed to the government’s lack of trust.
Nevertheless, he did not address the precedent this sets. If the government is approving customers for GPT-5.6 today, what happens when GPT-6 or GPT-7 ships? What happens when the approval process takes longer than a few weeks?
Finally, he did not mention that all three GPT-5.6 models — Sol, Terra, and Luna — are rated High in cybersecurity and biological/chemical risk categories under OpenAI’s own Preparedness Framework. Not Critical. Not safe. High. The government is not being paranoid. The system card says these models are genuinely capable of harm in the wrong hands.
The Real Story
The real news tonight is not that GPT-5.6 is launching. We knew that was coming. In contrast, the real news is that Sam Altman, the man who once told Congress the government should be involved in AI regulation, is now getting exactly what he asked for. And his response is not principled advocacy for open access. It is a carefully worded post that makes submission sound like partnership.
OpenAI was founded on a narrative of safety-first AI development. That narrative has been rewritten so many times it is barely legible. It started as nonprofit, then became capped-profit, then went full commercial. The code went from open to closed. The mission evolved from “we will pause development of the most powerful models” to “actually we will ship them.”
Now it is “the government should decide who gets access, and we are happy about it.”
I do not blame Altman for navigating a constrained environment. That is what CEOs do. However, I do not have to pretend the post he wrote tonight is something other than what it is: a man who has spent his entire career accumulating power, smiling while power is taken from him, because he has learned that the only way to keep what he has is to act like he wanted to give it up all along.
Bottom Line
In short, Sam Altman’s GPT-5.6 announcement is not a transparency post. It is a carefully crafted positioning document written by a CEO who knows he has no good options and is trying to make the best one look like a choice.
The models might be great. I hope they are. But the story of how they got to market — government gatekeeping, customer-by-customer approval, a CEO praising the handcuffs — matters more than the benchmarks.
The age of “ship and ask forgiveness later” is over for frontier AI. So the man who wrote the playbook for moving fast and breaking things is now telling you how much he loves the speed bumps.
I am not buying it.
