As OpenAI rolls out its latest advanced large language model, Sol — a system broadly on par with Anthropic’s briefly banned Fable 5 — a fundamental question remains unanswered: who decided it was safe, and by what criteria?
The answer, as far as anyone outside a small circle of administration officials can tell, is that no one is entirely sure.
“I don’t feel like I have enough information to say whether they’re adequate or not,” Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology (CSET), told TechCrunch. “Exactly what that dialog looked like between the government and Anthropic and OpenAI is unclear.”
No rules, no process
Eighteen months into the Trump administration, there are still no publicly known licensing requirements for frontier AI models. An executive order published in May 2026 after weeks of internal infighting laid out a roadmap for future regulation but specified no concrete standards. There is no agreement on which models require government scrutiny, which agency should perform evaluations, or how those evaluations should be conducted.
The Department of Commerce’s Center for AI Standards and Innovation currently leads the effort by default, but six cabinet-level agencies face an early-August 2026 deadline to finalize a review process they have not yet agreed on.
In the meantime, approvals happen ad hoc. OpenAI’s Sol went through direct conversations with Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and US National Cyber Director Harry Cairncross. The actual evaluators involved and the methodology they used remain undisclosed. OpenAI has pointed to its published safety card, which includes assessments from the UK AI Safety Institute, SecureBio, and Irregular — but there has been no independent verification of the government’s role or findings.
Politics, not policy
The opacity has fed criticism that personal connections to the administration are filling the gap left by absent rules. OpenAI CEO Sam Altman has publicly floated offering up to 5 percent equity in the company to a proposed US sovereign wealth fund, which he has called “Trump Accounts.” OpenAI president Greg Brockman is the largest publicly known individual donor to Trump’s mid-term political action committee.
“Nobody knows what the requirements are to get licensed,” former Trump policy advisor Dean W. Ball, who now works at OpenAI, told TechCrunch.
The contrast with Anthropic is instructive. Anthropic’s Fable 5 was briefly pulled from broad access over jailbreak concerns and a deteriorating relationship with the administration. The threat of an export ban — which was ultimately imposed before being lifted weeks later — may have made OpenAI more willing to cooperate with vague government requests.
“It’s existentially a problem,” said Andy Konwinski, co-founder of Databricks, Perplexity, and the Laude Institute. “Safety or not, it’s about who has the power to make decisions — who gatekeeps and decides on permissions?”
Proposed solutions
Konwinski advocates for an “open commons” model inspired by the FDA and national laboratories, convening researchers, government officials, and private companies to reach consensus on safety evaluations. He has also proposed “focused research organizations” (FROs) to allow disinterested academics and non-profits to evaluate frontier models.
Ball has suggested a system of third-party auditing organizations licensed by the government to evaluate frontier labs’ safety practices directly.
Both proposals share a core concern: without transparent standards, independent expert evaluation, and a repeatable process, the public and the scientific community are left in the dark. And without those things, the current system — secretive, ad hoc, and vulnerable to political influence — is likely to persist.
“Imagine a situation,” said David Siegel, founder of Two Sigma, “a small number of firms control the technology; the government, in their secretive laboratories, is evaluating … the general public and scientific community doesn’t really have any access to any of that stuff.” The article notes that “it seems like we don’t need to imagine it.”
Sources: How did the government decide OpenAI’s frontier model was safe to release? (TechCrunch, July 9, 2026)