OpenAI is deploying its latest advanced LLM, Sol, for broad public access. Sol is considered to be at least equivalent to the Anthropic fable. The Anthropic Fable is a model whose ability (or ownership) highlighted the White House and temporarily off-limits to the public.
So how did these models get the OK for release? Short answer: no one knows exactly.
“Frankly, I don’t know their exact process, so I certainly don’t feel like I have enough information to know whether they’re appropriate or not,” Mina Narayanan, senior research analyst at Georgetown’s Center for Security and Emerging Technologies, told TechCrunch. “Although Anthropic did state that it was in dialogue with the government, developed classifiers to detect jailbreak attempts, and put in place a defense-in-depth strategy to prevent future jailbreaks, it is unclear what exactly the dialogue between the government and Anthropic and OpenAI was.”
“No one knows what the requirements are to obtain a license,” Dean W. Ball, a former Trump policy adviser who now works at OpenAI, wrote in a newsletter last month.
Andy Konwinski, a computer scientist who co-founded Databricks, Perplexity and the Laude Institute, said he hasn’t talked to anyone who understands the process, even at Frontier Labs. “It’s an existential problem,” he told TechCrunch. “Being safe depends on who has the power to make decisions. Who is in control and who decides what is allowed?”
Eighteen months into the Trump administration, there is still little clarity on how to move forward, even though industry numbers are shaping policy, or so some critics claim. Last month, after weeks of infighting, an executive order was released laying out a roadmap for evaluating the Frontier model, but details other than those that don’t exist have yet to be finalized. “There will not be an FDA for AI,” Sriram Krishnan, a former partner at Andreessen Horowitz and until last month a senior adviser on AI at the White House, told the Financial Times.
Notably, there is still no agreement on what kinds of models require government scrutiny or which agencies should conduct those assessments. For now, the Commerce Department’s AI Standards and Innovation Center appears to be taking the lead, but the executive order directs six Cabinet agencies to finalize the process by early August. What has emerged in the meantime is, at best, ad hoc.
OpenAI CEO Sam Altman told CNBC that the process included conversations with officials such as Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and U.S. National Cyber Director Sean Cairncross, but it was not clear who the experts tested the model or how they tested it. OpenAI declined to share details of the government’s process with TechCrunch, but noted that the latest model’s safety card includes the results of several external assessments by organizations such as the UK’s AISI, SecureBio, and Irregulator.
Similar to Anthropic’s Fable rollout, OpenAI previewed the model for the government and a select group of users ahead of wider release, but it’s unclear who all of those users are and how they were selected. “We don’t believe this type of government access process should be the long-term default,” the company said in a blog post in late June, and said it would work with the government to pursue an alternative path.
But the context for these conversations includes Altman reportedly offering up to a 5% stake in OpenAI stock for the administration’s so-called “Trump Account,” and OpenAI President Greg Brockman’s role as the largest publicly known donor to Trump’s mid-term political operation. For outside observers, it is difficult to distinguish between these activities and the government’s apparently light-touch approach to regulating Sol.
Meanwhile, Anthropic’s Fable was temporarily removed from widespread access after the U.S. government banned its use by foreigners. That’s partly because of serious concerns about users jailbreaking this model to access hacking capabilities, and partly because of personality conflicts between Anthropic and the Trump administration. The threat of an export ban may also have made OpenAI more cooperative with (unknown) government requests.
From an industry perspective, a hands-off approach to regulation may be good, but an approach that relies on personal connections with government officials creates uncertainty and bad incentives.
Konwinski told TechCrunch that he’s concerned that the true experts in the technology — “not just safety researchers, alignment researchers, interpretability researchers, but also data people and people across the stack” — aren’t playing a sufficient role in the model release process.
Konwinski argues that “open commons” is the best way to actually achieve a balance between safety and innovation. He points to models like the FDA, NIH, and national laboratories that convene researchers, government officials, and private companies to reach consensus on safety issues.
Part of that is due to the capitalist incentives that have motivated AI researchers for more than a decade, and played out in court during Elon Musk’s lawsuit challenging OpenAI’s corporate structure. Ball points out that the nature of the AI business means that companies need to recoup much of their training costs as soon as a model is released and stay ahead of their competitors.
“Even if their intentions are well-intentioned, there are very clear legal obligations and fiduciary duties built into their business procedures,” Konwinski points out.
In his post, Ball argued that the way forward will depend on a government-approved third-party auditor evaluating Frontier Labs’ approach to safety. Konwinski is also bullish about new organizational forms, such as centralized research organizations, that can help more disinterested professionals in academia and nonprofit organizations access and evaluate frontier models.
For now, the secrecy surrounding the development of AI remains, but it will also pose a political challenge for an industry that Americans increasingly view with skepticism. “There’s no sense that responsible people are driving these changes,” Remzi Alpaci-Dussault, a computer science professor at the University of Wisconsin-Madison, said at the Open Frontiers conference last week.
At the same event, computer scientist David Siegel, who founded Two Sigma, one of the most successful quantitative hedge funds, asked attendees: “Imagine a situation where a small number of companies control the technology, and the government evaluates the technology in secret labs to see if it’s suitable for use, and the general public and the scientific community don’t really have access to it. I think that’s a very bad situation.”
It seems there is no need to imagine.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
