How did the government decide that the OpenAI boundary model was safe to release?

OpenAI is launching its latest LLM development, Sol, for wider public access. Sol is considered at least the equivalent of Anthropic’s Fable, a model whose capabilities (or ownership) impressed the White House enough that it was temporarily banned from public access.

So how did these models find the right fit for release? Short answer: No one is sure.

“Frankly, I don’t have visibility into those exact processes, so yeah, I don’t feel like I have enough information to say whether it’s enough or not,” Mina Narayanan, senior research analyst at Georgetown’s Center for Security and Emerging Technology, told TechCrunch. “Anthropic said they were in talks with the government, and that they were building a logger to detect jailbreak attempts, and that they had implemented loophole protection techniques to prevent future prison breaks, but what that conversation looked like between the government and Anthropic and OpenAI is unclear.”

Dean W. Ball, a former Trump policy adviser who now works for OpenAI, wrote that “no one knows what the requirements are to get a license” in his newsletter last month.

Andy Konwinski, a computer scientist who founded Databricks, Perplexity, and the Laude Institute, said he had never spoken to anyone who understood the process, not even frontier lab workers. “It’s an existential problem,” he told TechCrunch. “Security or not, it’s about who has decision-making power—who guards the gates and decides on permits?”

Eighteen months into the Trump administration, there is still little clarity about how to move forward, despite—or, some critics say, because of—the policy of imposing industry statistics. Last month, after weeks of fighting, an executive order was published setting out the guidelines for testing border models, but the details have not been filled in, except that there will be none. “There won’t be an FDA for AI,” Sriram Krishnan, a former partner at Andreesen Horowitz who served as a senior adviser on AI at the White House until last month, told the Financial Times.

Notably, there is still no agreement on what types of models require government scrutiny, or which agency or agencies should conduct that scrutiny. For now, the Department of Commerce’s Center for AI Standards and Innovation appears to be leading the way, but the executive order directs six cabinet agencies to decide on the final process by early August. What has emerged so far is, at best, ad hoc.

OpenAI CEO Sam Altman said on CNBC that the process includes discussions with officials such as Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and US national cyber director Sean Cairncross, but it is not clear who the experts are testing the models or how they did it. OpenAI declined to share details about the government’s process with TechCrunch, but pointed to the results of several external tests conducted by organizations such as UK AISI, SecureBio and Irregular on the latest model’s security card.

Like Anthropic’s Fable release, OpenAI previewed the government model and selected users before the wider release, but we don’t know who all those users were or how they were selected. In a late June blog post, the company said “we don’t believe this type of government access process should be a long-term mistake,” and said it would work with the government to develop a different way forward.

The background to those discussions, however, includes Altman reportedly donating up to 5% to OpenAI’s so-called “Trump Accounts,” and OpenAI president Greg Brockman’s role as the most publicly known major donor to Trump’s midterm political campaign. It is difficult for outside observers to separate those activities from the government’s seemingly simple way of controlling Sol.

Amthropic’s legend, on the other hand, was briefly removed from wide access when the US government banned its use by foreigners, partly due to genuine concerns about users breaking the model to access hacking capabilities and partly due to a personality conflict between Anthropic and the Trump administration. The threat of export bans may also lead OpenAI to cooperate more with (anonymous) government requests.

From an industry perspective, a deregulation approach may be good, but one that relies on personal contact with management creates uncertainty and negative incentives.

Konwinski told TechCrunch that he worries the real experts in this technology—”security researchers, alignment researchers, interpretation researchers, but also data people, people from all stacks”—are not playing enough of a role in the model release process.

Konwinski argues that the “open commons” is the best way to balance security and innovation. He points to models like the FDA, NIH, or national labs, which convene researchers, government officials, and private companies to reach consensus on safety issues.

Some of this comes down to the capitalist motivations that have motivated AI researchers for more than a decade, and played out in the courtroom during Elon Musk’s trial challenging OpenAI’s business model. Ball points out that the nature of the AI ​​business requires companies to recoup most of their training costs soon after their models are released and ahead of the competition.

“Even if their intentions are good, there are very clear legal obligations and fiduciary obligations built into the operating procedures,” notes Konwinski.

Ball, in his position, said the way forward would depend on third-party auditing organizations, licensed by the government, to assess the security of the labs at the border. Konwinski, too, is positive about new institutional formats such as focused research organizations that can help more disinterested professionals from higher education institutions and the nonprofit world to explore frontier models.

For now, the secrecy around AI development is not going away, but it will also create political challenges for an industry that Americans continue to view with skepticism. “There is no sense that responsible people are driving these changes,” University of Wisconsin-Madison computer science professor Remzi Arpaci-Dusseau said last week at the Open Frontier conference.

At the same event, David Siegel, the computer scientist who founded Two Sigma, one of the most successful hedge funds, asked attendees to “imagine a scenario, which I think is going to be worst, [where] a small number of firms that control technology; the government, in their secret laboratories, tests whether the technology is ready for use or not; and the general public and the scientific community don’t really have access to any of those things.”

It seems that there is no need to think about it.