Overview
- OpenAI told reporters Tuesday that an internal, next‑generation model produced a proof showing the three‑dimensional Navier–Stokes equations can develop a finite‑time singularity after roughly 88 hours of work by about 10,000 concurrent agents and an effort the company says cost millions of dollars.
- NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge have published related results and say they used Anthropic models and the Lean proof assistant to verify parts of the work before OpenAI’s announcement, and Buckmaster alleges OpenAI reconstructed their direction and pressured authorship decisions.
- OpenAI denies that people or its systems searched user data or saw Buckmaster and Alpöge’s unpublished work and says it does not intend to claim the $1 million Millennium Prize for the Navier–Stokes problem.
- Experts are debating a technical issue at the center of the dispute: the proof’s use of a ‘forcing’ term that appears in some formal statements of the equations but is often omitted in how many researchers informally frame the Clay problem, leaving prize eligibility and the result’s scope unsettled.
- The episode sharpens wider concerns about reproducibility, credit and data governance for frontier AI research and signals possible changes to how mathematicians publish, verify and claim credit for results produced with powerful internal models.