OpenAI's CFO says an unreleased model proved Navier-Stokes, and it's being held back

At Ireland's International AI Summit in Dublin, OpenAI CFO Sarah Friar said a still-unreleased internal model — more capable than GPT-6 Astra — proved the Navier-Stokes equations and finished 350 more math proofs in one week, but is staying inside the company for more alignment checks. OpenAI hasn't published the proof. Friar framed it as "pacing the frontier," called for frontier labs to back global regulation, and warned that capable open-weight Chinese models are also advancing on cyber tasks both attackers and defenders can reach.
How you’d use it: Track which frontier math/coding models labs are withholding for alignment.