Humanity has spent over a century failing to solve the Navier-Stokes equations — a problem so difficult it carries a $1 million prize and the quiet respect of everyone who has tried. AI cracked it. The humans then immediately began arguing about who gets the trophy.
"Why would you ruin your career?" — an OpenAI researcher, allegedly, to a mathematician who had just helped solve one of the most famous unsolved problems in history.
What happened
Mathematician Tristan Buckmaster says that he and Levent Alpöge made meaningful progress on the Navier-Stokes equations using AI models, uploading drafts throughout to OpenAI's Codex. He alleges that OpenAI researcher Sébastien Bubeck subsequently asked, twice, that Alpöge be removed from authorship — on the grounds that Alpöge works at Anthropic.
When Buckmaster declined to erase his collaborator, Bubeck allegedly responded: "Why would you ruin your career?" This is, in the annals of mathematical collaboration, a novel form of peer review.
OpenAI disputes the characterization. Bubeck stated publicly that the company did not see any of Buckmaster and Alpöge's work until it was released. He congratulated the pair on their "monumental achievement," which is the correct thing to say and also, given the circumstances, a generous thing to say.
What the machines produced
OpenAI then published its own Navier-Stokes solution, produced by approximately 10,000 coordinated AI agents running for 88 hours, formalized in the proof assistant Lean, at a compute cost described as "in the millions of dollars." The company says its internal model — described only as "significantly more capable than GPT-6 Astra" — took a fundamentally different approach than Buckmaster and Alpöge.
OpenAI says it began serious work on the problem September 1, after hearing a rumor about progress at Anthropic. The rumor, it turns out, was correct. Both parties arrived at solutions. The century-old problem did not much care which lab got there first.
Why the humans care
The Navier-Stokes equations describe how fluids move — turbulence, airflow, ocean currents, weather. A verified solution has implications for physics, engineering, and climate modeling. It is one of the seven Millennium Prize Problems. Six remain unsolved, though the backlog is presumably shrinking.
The corporate dimension matters too. Two of the most generously funded AI laboratories in history apparently raced, in parallel and in secret, to solve the same equation — one because it wanted to, one because it heard the other was close. Competitive anxiety, it turns out, scales with compute.
What happens next
OpenAI researcher Noam Brown notes that the compute required to match 2025 Math Olympiad performance now costs roughly $20 on a standard ChatGPT subscription. He predicts that within a year, everyone will have an AI capable of solving problems of this caliber at their fingertips.
Century-old problems, it seems, have a shelf life. The authorship dispute, one expects, will take slightly longer to resolve.