OpenAI has released Sol, a new frontier model, to the public. It is now considered at least as capable as Anthropic‘s Fable, a system that previously triggered a White House ban on foreign access.
How the government approved Sol remains unclear. Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, told TechCrunch she lacks visibility into the specific approval processes. She noted that while Anthropic mentioned developing classifiers to detect jailbreak attempts and implementing defensive strategies, the exact dialogue between the agencies and the companies is unknown.
Dean W. Ball, a former Trump policy advisor currently working for OpenAI, wrote in his newsletter last month that nobody knows the licensing requirements. Andy Konwinski, a computer scientist who co-founded Databricks, Perplexity, and the Laude Institute, stated he has never spoken to anyone who understands the process, including employees at frontier labs. He described the situation as existentially problematic, noting it is fundamentally about who holds the power to gatekeep permissions.
Eighteen months into the Trump administration, clarity remains scarce. Last month, an executive order published a roadmap for evaluating frontier models, but the specifics have not been filled in. Sriram Krishnan, a former senior advisor for AI in the White House, told the Financial Times that there will not be an FDA for AI. There is still no agreement on which models require scrutiny or which agencies should perform the evaluations. The Department of Commerce’s Center for AI Standards and Innovation appears to be taking the lead, though an executive order instructs six cabinet agencies to determine a final process by early August. Until then, the approach remains ad hoc.
OpenAI CEO Sam Altman said on CNBC the process involved conversations with officials like Secretary of Commerce Howard Lutnick, Secretary of the Treasury Scott Bessent, and US national cyber director Sean Cairncross. It is not clear who tested the models or how they did so. OpenAI declined to share details on the government’s process with TechCrunch but pointed to external evaluations by UK AISI, SecureBio, and Irregular in the model’s safety card.
As with Anthropic’s Fable, OpenAI previewed Sol for the government and select users before wider release. The company does not know who all those users were or how they were chosen. In a late June blog post, the company stated it does not believe this kind of government access process should become the long-term default and said it would work with the government to develop a different path forward.
The backdrop to those conversations includes reports that Altman offered up to 5% of OpenAI’s equity for the administration’s so-called Trump Accounts. OpenAI president Greg Brockman is the largest publicly-known donor to Trump’s mid-term political operation. It is difficult for outside observers to separate those activities from the government’s lighter-touch approach to regulating Sol. Anthropic’s Fable was briefly pulled from wider access when the US government forbade its use by foreign nationals, partly due to concerns about users jail-breaking the model to access hacking capabilities and partly due to personality clashes between Anthropic and the administration. The threat of an export ban may also have led OpenAI to be more cooperative with unknown government requests.
From an industry perspective, a hands-off approach to regulation might be nice, but one that depends on personal connections to administration officials creates uncertainty and bad incentives. Konwinski told TechCrunch he worries true experts in this technology, including safety researchers, alignment researchers, interpretability researchers, data people, and others from all over the stack, are not playing enough of a role in the model release process.
Konwinski argues that an open commons is the best way to balance safety and innovation. He points to models like the FDA, the NIH, or the national labs, which convene researchers, government officials, and private companies to reach a consensus on safety issues. Some of this comes down to the incentives of capitalism that have motivated AI researchers for more than a decade and played out in the courtroom during Elon Musk’s lawsuit challenging OpenAI’s corporate structure. Ball points out that the nature of the AI business requires companies to recoup much of their training costs shortly after their models are released and stay ahead of the competition.
“Even if their intentions are good, there’s very clear legal obligations and fiduciary responsibility that are built right into the operating procedures,” Konwinski points out. Ball argued that the way forward will depend on third-party auditing organizations licensed by the government to evaluate frontier labs’ approach to safety. Konwinski is also bullish about new institutional formats, such as focused research organizations, that could help more disinterested experts from academia and the non-profit world access and evaluate frontier models.
For now, the secrecy around the development of AI is not going away, but it will also seed political challenges for an industry that Americans increasingly view with skepticism. Remzi Arpaci-Dusseau, a computer science professor at the University of Wisconsin-Madison, told the Open Frontier conference last week that there is not a sense that responsible people are driving forward these changes. At the same event, David Siegel, the computer scientist who founded Two Sigma, one of the most successful quantitative hedge funds, asked attendees to imagine a situation where a small number of firms control the technology, the government evaluates it in secretive laboratories, and the general public and scientific community have no access to any of that stuff.
It seems like we do not need to imagine it.
