OpenAI Sol release: how the government cleared the frontier model

OpenAI is rolling out its latest advanced LLM, Sol, for wide public access, but the shape and participants of the government review that cleared it are opaque. The company previewed Sol for US officials and select users ahead of wider release, but declined to share details about who tested the model or how the testing was done.

How was Sol reviewed before release?

OpenAI previewed the model for the government and select users, and cited external evaluations such as U.K. AISI, SecureBio, and Irregular in its safety card, but the company would not disclose who the government’s testers were or the methods they used. OpenAI CEO Sam Altman said conversations happened with officials including Secretary of Commerce Howard Lutnick, Secretary of the Treasury Scott Bessent, and U.S. national cyber director Sean Cairncross, yet the specific experts and evaluation procedures remain undisclosed.

The lack of transparency extends to the mechanics of the preview. The company said in a late June blog post that it does not believe the current government access process should become the long-term default, and that it will work with officials to develop another path forward. OpenAI pointed TechCrunch to the results of several external evaluations in the model’s safety card instead of detailing the government testing process.

Who decides which models need scrutiny and who performs the checks?

There is no settled answer: the executive order published last month lays out a roadmap but leaves many specifics open, and it instructs six cabinet agencies to determine a final process by early August. For now, the Department of Commerce’s Center for AI Standards and Innovation appears to be taking the lead, but the order delegates responsibility to multiple agencies rather than naming a single regulator.

Industry figures and former insiders say that even people inside frontier labs do not understand the licensing pathway. Dean W. Ball, a former Trump policy advisor now at OpenAI, wrote that “nobody knows what the requirements are to get licensed.” Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, said, “Frankly, I don’t have visibility into those exact processes, so yes, I don’t feel like I have enough information to say whether they’re adequate or not.”

The Anthropic precedent highlights the variance in enforcement. Anthropic’s Fable was briefly pulled from wider access when the US government forbade its use by foreign nationals, partly over concerns about jailbreaks and partly because of personality clashes between Anthropic and the administration. That episode, combined with reports that Sam Altman offered as much as 5% of OpenAI equity for the administration’s so-called “Trump Accounts” and that OpenAI president Greg Brockman was a major donor to Trump’s mid-term operation, complicates the public’s ability to separate political ties from regulatory outcomes.

Why it matters

An ad hoc, connection-driven approval process concentrates power and raises incentives that could favor speed and market position over rigorous, independent safety evaluation. Experts quoted in the piece argue that true safety assessments require broader participation from safety, alignment, and interpretability researchers, plus data specialists across the stack. Some industry figures propose a model of licensed third-party auditors overseen by the government or institutional formats that give disinterested experts better access to frontier models.

The present secrecy also creates political risks for companies and for public trust. The article notes that Americans increasingly view the industry with skepticism, and it quotes voices who fear a future in which a small number of firms control the technology while secretive government labs evaluate it with little input from the public scientific community.

What to watch

Watch for the six cabinet agencies to publish their agreed process by early August, as the executive order requires. Also watch whether the Department of Commerce’s Center for AI Standards and Innovation formalizes a lead role, and whether the government or industry moves to license third-party auditors or new institutional formats for model evaluation.