The million-dollar math prize just became a dispute about provenance. On September 8, 2026, OpenAI announced that its agents had solved the Navier–Stokes Millennium Prize Problem, one of seven Millennium Prize Problems, by proving that the full equations can break down.
That breakdown means the equations can predict a fluid’s speed becoming infinite. OpenAI obtained the proof with an internal model that outperforms the Astra model, then certified the result using the programming language Lean. OpenAI does not plan to claim the one million dollar prize.
The claim matters because only one other Millennium Prize Problem had been solved before today. It also arrives with an awkward question attached: did OpenAI’s system reach the result on its own, or did it follow a path developed by mathematicians already working on the problem?
A proof arrives after a parallel effort
Tristan Buckmaster, an NYU mathematician, and Levent Alpöge, a mathematician at Anthropic, had spent almost a year working on the problem with publicly available models from OpenAI and Anthropic. Their work focused on blowing up the Euler equations, a method Buckmaster later claimed OpenAI had adopted.
On Monday, Buckmaster posted a proof on Mastodon showing that a simplified version of the Navier–Stokes equations can break down. The night before OpenAI announced its proof, Buckmaster alleged that OpenAI had learned about his and Alpöge’s progress during the last week and had adopted the same method.
OpenAI denied that any agents or employees accessed Buckmaster and Alpöge’s transcripts. Sébastien Bubeck, a member of OpenAI’s technical staff, put the company’s position plainly: “We did not use their prompt or models or proof.”
The method is older than the controversy
The approach used by the AI was called “forcing.” Diego Córdoba and Luis Martínez-Zoroa originally developed it, so the central mathematical idea did not appear from nowhere when OpenAI’s internal model produced its proof.
Córdoba said, “We’re a little bit in shock.” That reaction captures the strange shape of the announcement: OpenAI presented a proof of the full equations after Buckmaster posted a result for a simplified version, and the company said its proof had been developed over the weekend.
Those details do not settle who contributed what. They do define the dispute. Buckmaster and Alpöge had used one route to make the Euler equations blow up; OpenAI presented a result showing that the full Navier–Stokes equations can break down as well, while denying access to their work.
Lean certification gives the proof a formal check, but it does not answer the question of intellectual origin. A verified proof can establish that a mathematical statement follows from its steps; it cannot, by itself, establish who first found the method or whether another team’s work shaped the route.
That distinction now sits beside OpenAI’s claim that its agents solved the problem. The announcement is not only about whether an AI system can produce advanced mathematics. It is also about how mathematicians will assign credit when models, private systems, public tools, and human researchers work on the same question.
Javier Gómez-Serrano, a mathematics professor at Brown University, Mark Chen, OpenAI’s chief research officer, and Luis Silvestre, a mathematician at the University of Chicago, are among the names tied to the surrounding mathematical and OpenAI discussion. Their presence does not erase the core uncertainty: the proof may be certified, while its path remains contested.
OpenAI’s refusal to claim the one million dollar prize removes one obvious incentive from the story, but not the larger stakes. The seven Millennium Prize Problems were designed to mark the limits of modern mathematics; now one of those limits has become a test of how AI-generated research is checked, credited, and trusted.
The proof is the headline. The provenance fight is the forecast.
Based on
