AI might have produced its biggest breakthrough yet.
OpenAI has reportedly produced a proof cracking one of the most famous unsolved problems in mathematics — the Navier–Stokes equations — using an enormous fleet of AI agents, according to reports from Axios and Wired.
Per Axios, OpenAI began its push for a solution on September 1, after researchers inside the company heard rumors that two Millennium Prize problems had been solved. The outlet reported that roughly 10,000 concurrent agents worked on the Navier–Stokes effort, reaching a result in about 88 hours.
The claim, if it holds up, would be a landmark moment in AI-assisted science — and it lands in the middle of an increasingly bitter public fight over credit for the recent string of AI-driven breakthroughs in fluid dynamics, which broke today.
What OpenAI actually claims to have proved
A critical caveat sits at the heart of this story. The reported OpenAI proof concerns finite-time blow-up for the forced Navier–Stokes equations — a modified version of the problem that adds an external “forcing” term to the equations. That is not, by itself, the exact formulation attached to the Clay Mathematics Institute’s $1 million Millennium Prize, which asks whether solutions to the standard 3D incompressible Navier–Stokes equations remain smooth forever or can blow up on their own.
NYU mathematician Tristan Buckmaster — whose own AI-assisted work with Anthropic’s Levent Alpöge has been at the center of this saga — has said he was told an internal OpenAI model produced a roughly 100-page proof of blow-up for forced Navier–Stokes, following essentially the same technical route (through smooth forcing, in the tradition of the Córdoba–Martínez-Zoroa program) that he and Alpöge had been pursuing quietly for most of the past year.
As of this writing, OpenAI has not publicly released the proof, and it has not been independently verified. Even a genuine solution would face a long road to the Clay prize: the Institute requires publication, at least two years of general acceptance by the mathematical community, and a recommendation from its advisory board before any money is awarded.
The rumor mill that started it all
The September 1 start date points to just how much of this story has been driven by competitive rumor. The talk that “two Millennium Prize problems had been solved” traces back, at least in part, to viral social media posts claiming Anthropic’s Claude had cracked Navier–Stokes — claims that began as one X user’s prediction rather than any confirmed result.
Those rumors, in turn, were fueled by the real, verified work Buckmaster and Alpöge had been doing — which they were then forced to rush into publication before outside pressure, as Buckmaster told it in his explosive statement, accusing OpenAI of trying to muscle in on the discovery and erase Alpöge from the credit.
The credit fight, so far
The blowback has been swift. Sébastien Bubeck — the OpenAI scientist named in Buckmaster’s account, best known for co-authoring the 2023 “Sparks of AGI” paper — responded publicly, calling the allegations “false and inflammatory” and promising more to say.
Fields Medalist Terence Tao, meanwhile, has called Buckmaster and Alpöge’s own AI-assisted, Lean-verified proofs “a remarkable achievement” and suggested the work could genuinely help solve Navier–Stokes — while separately warning that a world where AI produces unreadable proofs the community can’t parse would be “a disaster” for mathematics.
Why it matters
Strip away the drama, and the business story here is remarkable either way. OpenAI reportedly spun up 10,000 parallel agents and turned around a frontier mathematical result in under four days — a pace of R&D that would have been unthinkable two years ago, and a direct signal of how the AI race is shifting from coding and chat toward raw scientific discovery. The episode also raises thorny questions the industry has barely begun to grapple with: who gets credit when an AI model produces a proof, how results get verified when no human can fully read them, and what happens when a viral rumor can push the world’s most valuable AI labs into a sprint for a million-dollar prize.
Whether OpenAI’s proof survives expert scrutiny — and whether it counts as “solving” Navier–Stokes at all — remains an open question. But the contest it has ignited, between OpenAI, Anthropic, and the mathematicians caught in the middle, is already one of the defining stories of the AI era.