OpenAI Claims a Navier-Stokes Breakthrough, but the Paper Trail Is Messier Than the Math

OpenAI announced on September 8 that an internal, unreleased system had produced what it called a resolution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems carrying a $1,000,000 award from the Clay Mathematics Institute since May 24, 2000. According to OpenAI's announcement, the system generated both an analytical proof and a Lean formalization showing that an initially smooth fluid at rest can develop a singularity in finite time.

The Clay Mathematics Institute has not accepted the claim. As reported by implicator.ai, the Institute's public problem listing still describes Navier-Stokes as unsolved, and Charles Fefferman of Princeton, who wrote the Institute's official problem description, told Quanta Magazine he was 'thrilled that the problem was solved' while noting the verification process has barely begun. Forkast News has gone further, arguing that OpenAI's forced result does not satisfy the criteria the Clay Institute originally set, meaning the prize itself remains unclaimed regardless of how the proof is characterized.

The timeline matters as much as the mathematics. Per Simon Willison's writeup, OpenAI's agents arrived at their result on Saturday, September 5, roughly 88 hours after the first agents were launched, with Lean verification following shortly after. That launch came, according to OpenAI researcher Sebastien Bubeck as reported by the-decoder.com, after the company began training a new model with advanced math capabilities on August 28, spurred by rumors that Anthropic-affiliated researchers had already resolved a major open problem.

Those researchers were Tristan Buckmaster, an NYU mathematics professor, and Levent Alpöge, a mathematician who works for Anthropic. Their own account, described in a PDF Buckmaster circulated and summarized on kingy.ai, dates their Boussinesq and Euler breakthrough to August 15, 2026, with Lean verification completed on August 22, after nearly a year of work making heavy use of Claude and Codex, particularly GPT-5.6 Sol. OpenAI's own account, published alongside its announcement, states that after completing its project and Lean verification on September 6, and hearing a rumor that Buckmaster and Alpöge also had a solution, the company reached out to coordinate release.

That coordination attempt is where the dispute ignites. Buckmaster says he was given two options, according to Glitchwire's reporting: agree to a joint announcement timed to OpenAI's release the next day, or publish immediately on his own. He chose to publish. Officechai reports that Bubeck later posted his own account acknowledging an 'ill-chosen remark made in frustration,' the now-circulated line 'why would you risk your career,' while disputing that it amounted to pressure, and noting that Buckmaster represented Anthropic's side throughout since Alpöge himself stayed largely out of the exchange.

Both teams, notably, built on the same prior mathematics. Quanta Magazine reports that both AI-enabled groups relied heavily on earlier work by Diego Cordoba of the Institute for Mathematical Sciences in Madrid and Luis Martinez-Zoroa, now at CUNEF University, whose doctoral research developed methods later adapted for both proofs. That shared foundation raises the question Willison poses directly in his post: if a researcher uses ChatGPT to make partial progress on a Millennium Prize problem, what are the odds that work quietly influences the training of a later model that then appears to solve the whole thing independently.

OpenAI, for its part, states plainly that neither its researchers nor its agents saw any of Buckmaster and Alpoge's work by any means before it was released publicly, and that no user data or private queries informed the system's output. Forkast News counters that OpenAI has not released its proof for independent verification, leaving mathematicians unable to audit the logic or the underlying reasoning traces that produced it. Until that manuscript, or the Lean project behind it, is opened to outside scrutiny, the Navier-Stokes problem stays exactly where the Clay Institute's website currently lists it: unsolved.