OpenAI is rolling out its latest frontier language model, Sol, for wide public access, following a period in which rival Anthropic's comparable model, Fable, was briefly banned from public release over safety and ownership concerns. The article's central point is that no one — including researchers, former officials and even frontier-lab employees — seems to understand how the US government actually decided these powerful models were safe to release, raising questions about who holds decision-making power over AI deployment.

Eighteen months into the Trump administration, there is still no agreed framework for which models require scrutiny or which agency should evaluate them. A recent executive order set out a roadmap but left specifics unfilled, ruling out an "FDA for AI"; six cabinet agencies are meant to finalise a process by early August, with the Commerce Department's Center for AI Standards and Innovation currently taking the lead. In the meantime, the approach appears ad hoc: Sam Altman described informal conversations with cabinet officials, while OpenAI pointed to external safety evaluations but declined to detail the government's process. Critics note the lighter-touch treatment of Sol coincides with OpenAI figures' financial and political ties to the administration, and warn that a system based on personal connections creates uncertainty, sidelines genuine experts and produces bad incentives.