TechCrunch probes opacity of US government review process for OpenAI's Sol model

TechCrunch examines how OpenAI's new model, Sol, and Anthropic's Fable were cleared for release, and concludes that nobody is quite sure. Sol is being rolled out widely and is described as at least comparable to Fable, which the White House briefly kept from public access. Outside experts quoted in the piece say they lack visibility into what the government asked of the labs. Mina Narayanan of Georgetown's CSET says she cannot judge whether the processes are adequate, and Andy Konwinski, a co-founder of Databricks, Perplexity and the Laude Institute, says no one he has spoken to understands them.

The article describes the policy backdrop as unsettled. An executive order issued the previous month sketches a way to evaluate frontier models but leaves details open, and it tells six cabinet agencies to settle a final process by early August. There is no agreement on which models need scrutiny or which agency should do the testing, though the Commerce Department's Center for AI Standards and Innovation appears to lead for now. Sam Altman said on CNBC that discussions involved Secretary of Commerce Howard Lutnick, Treasury Secretary Scott Bessent and national cyber director Sean Cairncross. OpenAI declined to give TechCrunch details but pointed to external evaluations listed in the model's safety card.

TechCrunch also notes reported links between OpenAI leaders and the administration, and says Fable was briefly pulled after the government barred foreign nationals from using it. Konwinski argues for an "open commons" that brings researchers, officials and companies together, while Dean Ball favors government-licensed third-party auditors. OpenAI wrote in late June that this access process should not become the long-term default.