Story
September 11, 2026
OpenAI’s Math Win Sparks a Bitter Fight Over Who Found the Path First
OpenAI portrays its Navier–Stokes result as an independent demonstration of rapidly advancing AI, while Tristan Buckmaster and Levent Alpöge see a troubling test of whether frontier labs can compete with researchers using their own tools. The dispute is less about whether the mathematics matters than about credit, data and power in AI-driven science.
For nearly a year, NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had been pursuing related fluid-dynamics questions with OpenAI’s Codex and Anthropic’s Claude. On Monday, they published results on the forced Euler problem — an important step toward the Navier–Stokes Millennium Prize challenge.
Buckmaster says the mood changed after he learned word of their progress had reached OpenAI. The route his team had chosen, he argued, was unusually specific: “It is not the direction one arrives at in a few days by giving a model the problem statement.”1 He also alleged that OpenAI researcher Sébastien Bubeck pressed to exclude Alpöge from a proposed joint paper because of his Anthropic affiliation — an allegation OpenAI disputes.
OpenAI’s account starts on September 1, when it heard rumors that two Millennium Prize problems had been resolved. It then set roughly 10,000 agents on open problems, with the Navier–Stokes run reaching a proof after 88 hours and formal verification taking another 17. The company says the effort consumed about 130 billion output tokens on that problem alone.2
On Tuesday, OpenAI announced a proof that smooth three-dimensional fluid flow can develop a finite-time singularity, calling it a major advance for mathematics. Greg Brockman hailed it as “a major milestone for both AI and mathematics” and a sign of a coming scientific renaissance.
3
But the breakthrough immediately became a dispute over provenance. OpenAI says neither its researchers nor agents saw Buckmaster and Alpöge’s work before it was public, though it conceded it “cannot rule out” that de-identified data derived from use of its products improved its models.2 Buckmaster views that caveat as central, not incidental.
Sam Altman defended the team’s conduct, saying it had sought a joint release after initially believing the outside researchers had solved Navier–Stokes.
4 Critics were unconvinced: a repost amplified by Yann LeCun congratulated Buckmaster and Alpöge while sarcastically praising OpenAI for not letting “any good customer transcripts go unmined.”
5
The result may yet stand as a landmark proof. But the episode has opened an equally consequential question: when labs supply the tools, training infrastructure and vast compute, who owns the intellectual trail to discovery?