OpenAI has released its latest advanced large language model, Sol, for public access. The model is considered at least on par with Anthropic’s Fable, a model that caused enough concern for the White House to briefly restrict its public use. Despite this, the exact process by which Sol was deemed safe for release remains unclear. 'Frankly, I don’t have visibility into those exact processes,' said Mina Narayanan, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, adding that the government’s evaluation criteria are not well understood. 'Nobody knows what the requirements are to get licensed,' noted Dean W. Ball, a former Trump policy advisor now working for OpenAI, in a recent newsletter. The lack of clarity has led to speculation about the role of personal connections in the approval process, with some observers suggesting that OpenAI’s more cooperative stance may have been influenced by its ties to the Trump administration.
The regulatory framework for evaluating frontier models remains underdeveloped, with an executive order published last month outlining a roadmap but leaving many specifics undefined. 'There will not be an FDA for AI,' said Sriram Krishnan, a former Andreesen Horowicz partner and White House AI advisor, highlighting the absence of a centralized regulatory body. The Department of Commerce’s Center for AI Standards and Innovation appears to be leading the effort, but six cabinet agencies are tasked with finalizing a process by early August. OpenAI declined to share details on the government’s evaluation process but cited external assessments by organizations like UK AISI and SecureBio as part of its safety evaluation.
The release of Sol follows a pattern seen with Anthropic’s Fable, which was temporarily restricted due to concerns over potential misuse. OpenAI previewed Sol for the government and select users before wider release, though the criteria for selecting those users remain unknown. The company has stated that it does not believe the current government access process should be the long-term standard, but the broader context includes reports of OpenAI offering equity to the administration for access to Trump’s political accounts. Critics argue that the lack of transparency and reliance on personal connections undermines the credibility of the regulatory process. 'There’s not a sense that responsible people are driving forward these changes,' said Remzi Arpici-Dusseau, a computer science professor at the University of Wisconsin-Madison, at the Open Frontier conference. Source: techcrunch
