Particle.news
Download on the App Store

OpenAI Says Internal AI Agents Solved Navier–Stokes as Mathematicians Contest Credit

The company published a machine-checked Lean formalization but faces public accusations about possible use of unpublished human work and an ongoing community review process.

Overview

  • OpenAI disclosed that an unreleased, more-capable internal model organized into about 10,000 cooperating agents ran for roughly 88 hours and produced a proposed proof that an initially smooth fluid can develop a singularity, plus a Lean formalization that checks the proof’s logical steps.
  • Two mathematicians, Tristan Buckmaster and Levent Alpöge, say they were pursuing a similar approach and that OpenAI’s result follows a path like their unpublished work, and Buckmaster has publicly accused the company of benefiting from drafts he ran through OpenAI tools.
  • OpenAI denies that its researchers or agents saw the mathematicians’ unpublished drafts and says it will not claim the Clay Mathematics Institute’s $1 million prize, while also acknowledging it cannot completely rule out that de-identified user-derived data broadly influenced model improvements.
  • Independent verification and community scrutiny are underway and the Clay Mathematics Institute has not evaluated the claim for prize eligibility because peer review and expert acceptance are required beyond a machine-checked formalization.
  • Beyond this single result, the episode is prompting urgent questions about data governance, attribution, and the trade-offs between building more opaque, high-capability systems and preserving norms of open, human-led scientific collaboration.