OpenAI previewed its GPT-5.6 model family, then launched it in limited preview to roughly 20 organizations approved by the U.S. government, at the government's request.
It is the first time an American company has released a frontier model under a government-managed access list, and Paul Roetzer says it marks the moment the soft nationalization of AI began.
OpenAI previewed its next-generation GPT-5.6 model family under a new naming system: Sol, the flagship model OpenAI calls its strongest yet; Terra, a balanced model for everyday work that delivers GPT-5-level performance at half the price; and Luna, a fast, low-cost option. The bigger story was not the models themselves but how they were released. Instead of a normal launch, OpenAI started with a limited preview for a small group of trusted partners, reportedly around 20 organizations whose participation was approved by the U.S. government. The models are available first through the API and Codex, OpenAI's coding agent, before any broader rollout.
OpenAI said it took this step at the government's request, previewing the models to the administration ahead of launch and agreeing to a phased release while it works with the White House on a cyber executive order framework. This traces back to a June 2 executive order asking AI companies to voluntarily give the government up to 30 days of pre-release access to models with advanced cyber capabilities. Reporting from The Information, Bloomberg, and Axios indicates Commerce Secretary Howard Lutnick personally called Sam Altman to warn against releasing GPT-5.6 publicly without government sign-off first.
It follows the saga where the Trump administration forced Anthropic to take its Fable 5 and Mythos 5 models offline after a reported jailbreak, the first time the federal government led to a commercial AI model going down. GPT-5.6 is the next milestone: the first time an American company has launched a frontier model under a government-managed access list.
OpenAI pushed back even as it complied, writing that it does not believe this kind of process should become the long-term default because it keeps the best tools from the users, developers, and enterprises who need them.
SmarterX founder and CEO Paul Roetzer broke down what it all means on Episode 222 of The Artificial Intelligence Show.
3 - Number of models (Sol, Terra, and Luna) in the new GPT-5.6 family
~20 - The number of organizations approved by the U.S. government for first access
30 - Voluntary pre-release access window (in days) from the June 2 executive order
11.3 hours - Length of a human coding task GPT-5.6 Sol can do fast at 50%-plus reliability, per evaluator METER
Government Oversight is happening. Asked whether the U.S. government is now going to decide which AI models get released, Roetzer was direct. "Until there's a more formal structure, it sure seems that way," he says. A very small group of businesses, plus the government, now has access to the most powerful models that the rest of society does not. Soft nationalization, meaning the government treats these models as strategic national-security assets without formally taking ownership, is what Roetzer sees.
"So what we could probably say with relative confidence is that soft nationalization has begun.
"This means not taking ownership and control yet, that these AI labs are still private companies, but the US government is increasingly treating their most capable models as strategic national security assets, not traditional like commercial software."
— Paul Roetzer, founder and CEO of SmarterX, Episode 222 of The Artificial Intelligence Show
The release decision is now political. A lab may believe a model is safe enough to release, but the government can step in and ask for a delay, narrow the access, or add safeguards. "The release decision becomes political. This is 100% happening," Roetzer says. In his view, access becomes tiered, with U.S. companies, agencies, and trusted partners first in line while the broader public and international access come later or never. Altman acknowledged the tension on X, writing that "at the request of the U.S. government, it is launching today in limited preview instead of the open access launch we were planning on." He added that a required preview period for extended red teaming — giving experts time to find a model's vulnerabilities — "is not a bad idea," but said, "I just don't like the idea of the government picking the customers."
A select few making the calls creates obvious risk. When a small group inside the government controls timing and access, the potential for insider trading and officials using their position for private financial gain (self-dealing) "becomes massive," Roetzer says. He points to White House staff being warned in March not to use non-public information to place bets on prediction markets such as Kalshi and Polymarket. The uncertainty is already moving markets. Per a June 25 New York Times report, OpenAI is now leaning toward holding off its IPO until next year.
This is not a partisan issue. As Roetzer notes, self-dealing and insider trading are realities of government regardless of which party holds office. The change is that the key asset is no longer chips or data centers. It is the model ability, the trained parameters that make a model work, plus API access and cyber capabilities, and the government is moving to control all of it. That shifts frontier AI from a product you buy into a capability the state rations.
And so far, there are no consistent rules. The government still has not told Anthropic what it must do to get Fable 5 back online. "Now businesses, American citizens, and our international allies are reliant on the whim of government with no clearly defined rules to determine if and when we get access to the frontier models," Roetzer says. For business leaders, that introduces a new kind of supply risk: the most capable tools might not be available to you. And no one can predict if and when they will be.
The China endgame defines the ceiling on how far this goes. The government faces a genuine bind. Move too slow or too closed, and the U.S. risks weakening its commercial AI lead. If officials decide frontier models are too powerful to release widely and hand the public the models that are a generation or two behind, Roetzer warns, "China can take the lead." That competitive pressure is the main force working against deeper control.
The atomic-weapon parallel. Roetzer compares the moment to Los Alamos in the 1940s, imagining a private company building the bomb while the government waited to take it seriously. It is an imperfect analogy because AI can do enormous good and lacks a weapon's visible destruction. But the labs believe these systems can be as disruptive to the economy and society as a weapon. "If you have now created something outside of the government that protects the stability of the democracy, how do you not take some level of control of it?" Roetzer asks. Whether Washington formalizes a transparent process or keeps making arbitrary calls is the thing to watch next.
While the government decides who gets the frontier models, most companies are nowhere near using it. Only 25% of organizations have reached the Scaling phase of AI adoption, according to the 2026 State of AI for Business Report. The largest share, 47%, is still Piloting, and 28% are still just trying to understand AI. The fight over restricted access to the most powerful models matters most for the small slice of organizations actually ready to deploy them.
The report is built on more than 2,100 responses across roles, functions, and industries, and it maps exactly where organizations sit on the path from understanding AI to scaling it. If you want a clear benchmark for where you stand before the access question even reaches your door, this is it. Read the full report →