OpenAI has rolled out its latest large language model Sol for public use, while Anthropic’s Fable model – touted as equally capable – sparked enough concern at the White House to briefly halt its public release. Both are classified as "frontier models," the most advanced AI systems to date, but the exact standards the U.S. government applied before granting clearance are still murky.

Opacity and Expert Frustration

Georgetown’s Center for Security and Emerging Technology senior analyst Mina Narayanan told TechCrunch, "I don’t have visibility into those exact processes, so I can’t say whether they’re adequate or not." She added that Anthropic said it was in talks with the government, had built a classifier to detect jailbreak attempts, and employed defensive gap strategies – yet the substance of those conversations with both Anthropic and OpenAI remains undisclosed.

Policy Made by Personal Connections?

Former Trump policy adviser Dean W. Ball wrote in his newsletter, "Nobody knows what the requirements are to get licensed." Similarly, Andy Konwinski – co‑founder of Databricks, Perplexity and the Laude Institute – confessed he has never spoken to anyone who truly understands the approval pipeline, even among frontier‑lab employees. "It’s existentially a problem," he told TechCrunch, "Safety or not, it’s about who has the power to gatekeep and decide on permissions?"

Executive Order, Yet No Concrete Rules

Eighteen months into the Trump administration, the policy landscape remains vague. A recent executive order outlined a roadmap for evaluating frontier models, but the specifics are still missing, leaving only what “won’t exist.” Former Andreessen Horowitz partner Sriram Krishnan, who served as a senior AI advisor to the White House, told the Financial Times, "There will not be an FDA for AI." No consensus exists on which types of models merit government scrutiny or which agency should conduct the reviews. For now, the Department of Commerce’s Center for AI Standards and Innovation appears to be leading, while six cabinet agencies have been tasked to finalize a process by early August – a process that, to date, is largely ad‑hoc.

OpenAI’s Government Dialogue

OpenAI CEO Sam Altman told CNBC that the process involved talks with officials such as Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and U.S. National Cyber Director Sean Cairncross. However, the identities of the experts who tested the models, and the methods they used, were not disclosed. OpenAI declined to share internal government‑review details with TechCrunch, instead pointing to external safety evaluations by UK AISI, SecureBio and Irregular, documented in the model’s safety card.

Industry Perspective and the Way Forward

From an industry standpoint, a hands‑off regulatory approach sounds attractive, but when approvals hinge on personal relationships with administration officials, uncertainty and perverse incentives arise. Konwinski warned that genuine safety researchers, alignment scholars, interpretability experts, and data specialists are not sufficiently involved in the release pipeline. He advocates an "open commons" model – akin to the FDA, NIH, or national labs – where researchers, government officials and private firms collaboratively set safety standards.

Ball and Konwinski both argue that third‑party auditors, licensed by the government, could provide the needed oversight. They also envision new institutional formats – focused research organisations that bring independent academic and nonprofit expertise into the frontier‑model evaluation process.

In short, while frontier AI development races ahead, the regulatory framework lags behind. The reliance on personal connections rather than transparent, systematic review creates a vacuum that could erode public trust and compromise safety. The industry’s call for structured, publicly accountable audits is louder than ever, and it may shape the next wave of AI governance.