Story
September 9, 2026
OpenAI’s Math Triumph Is Being Swallowed by a Fight Over Credit
OpenAI portrays its result as a landmark demonstration of rapidly advancing AI research, while Tristan Buckmaster and Levent Alpöge’s side sees a troubling test of whether mathematicians can safely use private lab tools without being outgunned or overlooked.
For nearly 90 years, the Navier–Stokes existence and smoothness problem has stood as one of mathematics’ most formidable unanswered questions: can equations describing fluid flow break down into a finite-time singularity? It is one of the Clay Mathematics Institute’s seven Millennium Prize Problems, each carrying a $1 million award.
The dispute began before OpenAI’s announcement. NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had spent almost a year using publicly available AI tools to investigate a related forced Euler result. Buckmaster said their drafts had been entered into OpenAI’s Codex, and questioned whether that work could have informed OpenAI’s parallel effort. He said the route they chose was not an obvious one to reach “in a few days by giving a model the problem statement.”1
OpenAI says it began its own push on September 1 after hearing rumors that two Millennium problems had been resolved. It deployed roughly 10,000 concurrent agents powered by an unreleased model, and says they found a Navier–Stokes proof about 88 hours later. The company’s write-up says the proof shows an initially smooth fluid can develop a singularity in finite time, and includes a Lean formalization.2
The clash came into public view after Buckmaster and Alpöge released their Euler work, followed by OpenAI’s full Navier–Stokes claim. Buckmaster alleged OpenAI pressed for an arrangement that would exclude Alpöge from authorship because of his Anthropic role; OpenAI and its chief executive, Sam Altman, deny that account. Altman said the company had initially sought a joint release and that its team acted “with integrity and generosity throughout.”
3
OpenAI insists neither its researchers nor agents saw the pair’s work before it became public, though it conceded it could not entirely rule out that de-identified product-usage data had helped improve its models. It also says the proofs and even their Euler results differ. The company will not claim the prize.
That leaves the larger question unresolved: if frontier labs can turn rumors into multi-million-dollar, 10,000-agent research campaigns, private AI assistants may become both mathematicians’ tools and their most powerful competitors.