OpenAI agrees to stagger rollout of its most powerful model to only Trump-approved customers

It is the second time in a month that a frontier lab's most powerful model has been held back from general release over fears about cyber capabilities.

OpenAI is staggering the rollout of its newest and most powerful AI model after a request from the Trump administration. To get access to the new model, customers must first be cleared by the U.S. government, the company said on Friday.

The model, called GPT-5.6 Sol, is the flagship in a new tier of more advanced models that includes a more efficient model called Terra and its cheaper cousin Luna. OpenAI says that Sol is its strongest model yet, able to complete 50% of long-running professional tasks and tops all previous OpenAI models on coding capabilities. OpenAI said it hopes to make all three generally available in the coming weeks.

The Information first reported that the Trump administration asks OpenAI to stagger release of the new model over security concerns.

The move represents a broader shift in how the U.S. government is approaching frontier AI. Advanced cyber capabilities displayed by Anthropic’s Mythos and OpenAI’s GPT-cyber have caused concern in Washington. By limiting access to the government is attempting to ensure that those capabilities don’t end up in the hands of bad actors or hostile nation-states

It is also the second time in a month that a frontier lab’s best model has been held back from general release over capability concerns. In early June, the Commerce Department issued export controls on Anthropic that forced the lab to cut off foreign access to two of its top models, citing national security concerns. Anthropic disputed the order, but was left with no choice but to pull the models offline.

Earlier this month, Trump also signed an executive order directing federal agencies to establish a framework under which AI companies could voluntarily provide the government with early access to powerful new models for up to 30 days before broader release. 

OpenAI describes its own situation as voluntary, in contrast to Anthropic’s situation.

“As part of our ongoing engagement with the U.S. government, we previewed our plans and the models’ capabilities ahead of today’s launch. At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government,” the company said in a blog post. 

However, the company also said it was not in favour of this kind of government access process becoming the “long-term default.”

We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks,” the company wrote, adding it was working with the Administration to develop the cyber Executive Order framework and a “repeatable process for future model releases.”

Capability concerns

OpenAI emphasized that Sol made its strongest gains in cybersecurity, specifically vulnerability and exploitation. There will be two new modes: “max,” and “ultra,” which will allow the model to reason longer and coordinate agents for specific tasks. On a key cybersecurity benchmark, OpenAI previously said the model was “competitive with” Anthropic’s Mythos. GPT-5.6 Sol uses approximately one third of the tokens used by Mythos but appears to lag slightly behind Mythos 5, a slightly more capable model from Anthropic.

OpenAI is pairing the release with what it calls its most extensive safeguards to date, and says that the model preview will police its own use. For higher-risk cases, the company says a larger model will review the conversation and could withhold responding if it’s judged to violate policy. 

It said that, despite the government gating, Sol did not cross the “Cyber Critical” threshold in its “Preparedness Framework”: in tests with Firefox and Chrome, it found the seeds of an exploit but did not produce a working one. OpenAI said it had spent 700,000 GPU hours hacking itself to try to identify vulnerabilities, and humans will conduct two more weeks of the tests before launch.

The limited rollout is a transitional period, and linked to President Trump’s June 2 executive order that directed agencies to build a framework for vetting models before release, according to OpenAI. Since that framework doesn’t exist yet, OpenAI says it conducted a phased rollout at the government’s request. 

The initial users are customers who have been approved by the US government, with the list expanding next week, according to OpenAI. The company said that the process looks like OpenAI sharing names and the government giving feedback. 

Sol is priced at $5 per million input tokens and $30 per million output tokens, compared to Terra at $2.50 and $15, respectively, and Luna at $1 and $6. 

An improvised licensing regime 

The recent steps toward any kind of attempt to regulate AI also represents a striking reversal for an administration that, on its first day in office, had rescinded a Biden-era requirement for AI companies to submit safety tests to the government, calling it overly burdensome.

However, critics have argued that, by pursuing an ad-hoc approach to containing the risks, what is emerging looks less like a coherent regulatory system and more like an improvised licensing regime. Jonathan Iwry, a fellow at the Wharton Accountable AI Lab, previously told Fortune that the government is “repurposing existing legal authorities into what is effectively a backdoor licensing regime.”

Dean Ball, a former Trump administration AI adviser who has since become a vocal critic of its recent decisions, argued that since Mythos, the United States has had an “informal” licensing regime for AI, “with no consistent rules or firm boundaries on state power or public transparency.”

Critics warn that an informal system, with no published criteria or appeal process, opens the door to discrimination—giving the government unchecked power to decide which companies get access to the market and which do not, with no legal recourse for those on the wrong side of that decision.

This story was originally featured on Fortune.com