Overview
- OpenAI disclosed on Tuesday that a coordinated multi-agent run of roughly 10,000 AI agents working about 88 hours produced an analytic proof it says shows the Navier–Stokes equations can develop a finite-time singularity.
- The company released a 165-page paper and a Lean proof formalization that it says was produced after the analytic result was translated and structured with its models to allow machine checking by experts.
- OpenAI says the effort used a new in-development model, Codex for idea synthesis, and GPT-6 Astra for Lean translation, and the company estimates the work required massive token volumes and millions in compute costs.
- Mathematicians Tristan Buckmaster and Levent Alpöge say they uploaded private drafts to Codex and voiced serious suspicion that those notes influenced the result; OpenAI denies specific access but says it cannot completely rule out de-identified training data having helped model development.
- Independent mathematical review and scrutiny of the Lean files are under way and will determine acceptance of the claim, and the dispute over data provenance could prompt new norms and audits for how researchers use AI tools.