TheModelDesk

OpenAI is launching its advanced large language model Sol, which is viewed as comparable to Anthropic's Fable. While Fable previously faced a brief public access ban due to White House concerns over its capabilities, both models are now being released to the public.

However, the criteria for granting such releases remain mysterious. Experts, including Mina Narayanan from Georgetown's Center for Security and Emerging Technology, note a lack of visibility into the government's approval process. While Anthropic mentioned developing jailbreak detection and defense-in-depth strategies, the specific nature of the dialogue between these companies and the government is unknown.

Industry figures have expressed concern over this opacity. Dean W. Ball, a former Trump policy advisor now at OpenAI, stated that licensing requirements are unknown. Computer scientist Andy Konwinski further argued that the lack of a clear decision-making framework and gatekeeping process presents an existential problem for the industry.

Despite an executive order from the Trump administration intended to provide a roadmap for frontier model evaluation, the document lacks concrete specifics. Sriram Krishnan, a former White House AI advisor, confirmed that the government will not establish an AI equivalent to the FDA. Currently, there is still no consensus on which specific models require scrutiny or which agencies are responsible for overseeing them.