AI
A Black Box Around AI Safety: How OpenAI Got the Green Light to Release Sol Remains a Mystery
Experts say the government's process for approving frontier AI models is opaque, potentially conflicted, and raises questions about who really holds power over these decisions.
TechCrunch — AI
OpenAI's latest model, Sol, is now available to the public—yet nobody can quite explain how the government decided it was safe to release. The approval process has become a black box, leaving researchers, technologists, and policymakers uncertain about what standards even apply.
Sol matches or exceeds the capabilities of Anthropic's Fable, a model the White House briefly restricted over safety concerns. Yet the paths these two systems took to market were strikingly different. That contrast has exposed a critical gap: the Trump administration lacks any formal, transparent framework for vetting frontier AI systems.
The Mystery Approval Process#
When asked about the mechanism for clearing models, experts drew blanks. Mina Narayanan, a senior research analyst at Georgetown's Center for Security and Emerging Technology, acknowledged she lacks visibility into how decisions are made. "Anthropic said they were in conversations with the government and developed tools to detect jailbreak attempts," she explained, "but exactly what that dialogue looked like is unclear."
Dean W. Ball, a former Trump policy advisor now working for OpenAI, was blunt on this point: "Nobody knows what the requirements are to get licensed," he wrote in a recent newsletter post.
Andy Konwinski, co-founder of Databricks and a computer scientist, said he's never met anyone—not even at frontier labs—who understands the process. "It's existentially a problem," he told TechCrunch. "Safety or not, it's about who has the power to make decisions and who gatekeeps."
A Government Still Finding Its Way#
Eighteen months into the Trump administration, clarity remains elusive. An executive order issued last month outlined a roadmap for evaluating frontier models, yet specifics remain unfilled. Sriram Krishnan, a former venture capitalist who recently stepped down as the White House's senior AI advisor, made clear there will be no formal FDA-style agency for AI.
The Department of Commerce's Center for AI Standards and Innovation is leading for now, but six cabinet agencies have until early August to settle on a final process. In the meantime, the system is entirely ad hoc.
OpenAI CEO Sam Altman mentioned conversations with Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, and National Cyber Director Sean Cairncross, but the company declined to reveal who tested Sol or how they evaluated it. The company pointed instead to external safety reviews from organizations like UK AISI and SecureBio—published on OpenAI's own safety card.
