OpenAI just pushed its most advanced model yet, Sol, out to the wider public. It’s at least as capable as Anthropic’s Fable, the model that spooked the White House enough to briefly yank it from public access. So how did either one get the green light? According to TechCrunch AI, nobody can actually say. That’s the whole story here, and it’s a bigger deal than it sounds.
What stands out is that the people closest to this admit they’re in the dark. “Frankly, I don’t have visibility into those exact processes,” Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, told TechCrunch AI. Dean W. Ball, a former Trump policy advisor now working for OpenAI, wrote that “nobody knows what the requirements are to get licensed.” Andy Konwinski, who co-founded Databricks and Perplexity, says he’s never met anyone who understands the process, including employees at frontier labs. When the insiders are confused, that’s not a detail. That’s the system.
What’s actually happening
Eighteen months into the Trump administration, there’s still no settled answer on which models need government review or who should run the tests. TechCrunch AI reports that an executive order last month laid out a roadmap, but the specifics are blank. What officials have confirmed is mostly what won’t exist. “There will not be an FDA for AI,” Sriram Krishnan, a former Andreessen Horowitz partner who advised the White House until last month, told the Financial Times.
For now, the Department of Commerce’s Center for AI Standards and Innovation appears to be leading, with six cabinet agencies ordered to nail down a real process by early August. Until then, the approach is ad hoc. Sam Altman said on CNBC the review involved conversations with the Commerce and Treasury secretaries and the national cyber director. Who tested the models, and how? Not disclosed. OpenAI pointed TechCrunch AI to external evaluations from groups like U.K. AISI, SecureBio, and Irregular, but declined to detail the government’s own role.
Why it matters now
The uncomfortable part is the backdrop. TechCrunch AI notes Altman reportedly offered up to 5% of OpenAI’s equity toward the administration’s “Trump Accounts,” and OpenAI president Greg Brockman is the largest publicly known donor to Trump’s midterm operation. Meanwhile Anthropic’s Fable got pulled from wider access, partly over real jailbreak concerns and partly, per the reporting, over personality clashes with the administration. When access depends on who you know and how you give, that’s not regulation. That’s a relationship.
For the industry, a light touch sounds great until you realize the rules can flip with a phone call. “It’s existentially a problem,” Konwinski told TechCrunch AI. “It’s about who has the power to make decisions.” He argues for an “open commons” modeled on the FDA, NIH, or the national labs, where researchers, officials, and companies reach consensus in the open. Ball sees the path running through government-licensed third-party auditors.
Where this goes next
Here’s my read on the next one to three years. Expect the August deadline to produce a framework that’s real on paper and negotiable in practice. Third-party auditing firms will emerge as the compromise everyone can live with, because no agency wants to own a launch that goes wrong. And the labs racing to recoup enormous training costs will keep pushing for speed, since fiduciary duty rewards shipping first.
What should you do about it if you build or buy AI?
- Don’t assume a model is vetted because it’s public. “Approved” right now can mean a few private meetings. Read the safety cards and the external evals yourself.
- Watch the August process. If licensed third-party auditors become the standard, that’s the credential to look for in your vendors.
- Build for regulatory whiplash. If your product depends on a specific frontier model, keep a fallback. Access rules here move on politics, not just capability.
- Track who’s setting policy. When industry figures write the rules they’ll be judged by, the incentives deserve your attention.
The technology is moving faster than the referees. For now, the honest answer to how Sol got cleared is a shrug. That won’t hold for long. More detail is in the original TechCrunch AI report.