How did they decide OpenAI frontier model was safe? (techcrunch.com)

OpenAI has announced the public release of its advanced language model, Sol, which is on par with Anthropic’s Fable—a model previously restricted due to concerns about its capabilities. The safety approval process for these models, however, remains unclear, raising questions about the transparency and accountability of regulatory measures in the AI/ML space. Experts have expressed confusion over the criteria and agencies involved in evaluating such frontier models, contributing to a perceived "existential" governance challenge where power dynamics and personal connections seem to overshadow rigorous safety assessments. The debate over AI model safety highlights significant implications for the industry. Current regulatory frameworks are ad hoc, lacking consensus on which models warrant scrutiny or which agencies should oversee evaluations. Critics argue that the informal, preferential access granted to select users and government officials creates an uneven playing field, threatening the broader development and trust in AI technologies. Stakeholders are calling for a more structured approach involving independent third-party audits and the participation of diverse experts to ensure that safety and innovation are balanced, rather than left to opaque processes potentially influenced by corporate interests.