![Navier-Stokes Is Solved. The Credit War Is Not.](https://cdn.slatesource.com/c/3/f/c3f860a4-0071-4451-b108-4cc92e742c82.webp) # Navier-Stokes Is Solved. The Credit War Is Not. - [Made in Slatesource](https://slatesource.com/@steph/navier-stokes-is-solved-the-credit-war-is-not) - By [Steph](https://slatesource.com/@steph) - Created on Sep 11, 2026 ## OpenAI solved a $1 million math problem in 88 hours. The fight over who owns that result is only beginning. On September 8, 2026, OpenAI announced that a swarm of roughly 10,000 autonomous AI agents had proved a singularity exists in the Navier-Stokes equations, resolving one of the six remaining Millennium Prize Problems posed by the Clay Mathematics Institute in 2000, each worth $1 million. Navier-Stokes governs how fluids move, from galaxy-scale turbulence to the wake behind a plane wing. The central question, open since 1822, is whether solutions can "blow up" to infinity in finite time. OpenAI's agents said yes. The proof took 88 hours, consumed an estimated several million dollars of compute, and was formally verified in Lean before the company went public. Within twelve hours of the announcement, NYU mathematician Tristan Buckmaster accused OpenAI of building its solution on work he and his collaborator Levent Alpoge of Anthropic had already done, without credit. ![](https://cdn.slatesource.com/c/3/f/c3f860a4-0071-4451-b108-4cc92e742c82.webp) ## The proof that made it possible The story does not begin with OpenAI. In 2023, Diego Cordoba (Institute for Mathematical Sciences, Madrid) and Luis Martinez-Zoroa (CUNEF University) developed new analytic techniques using "infinite cascades" of non-singular solutions. Their 2023 paper proved singularities exist in the Euler equations (a related but simpler fluid system) under a messy forcing term. They fell short of the Navier-Stokes Millennium criteria, which require a smooth forcing term. That gap, smooth versus messy, was the last kilometre of a century-long marathon. It was exactly where Buckmaster and Alpoge were working. Princeton mathematician Charles Fefferman, one of the world's leading authorities on Navier-Stokes, was unambiguous: "The heroes of the story are Cordoba and Martinez-Zoroa." Martinez-Zoroa, for his part, offered a dignified response: "It would have been nice to do this ourselves, but I'm very happy for him." [OpenAI: On the Navier-Stokes Millennium Prize Problem (primary source, September 8, 2026)](https://openai.com/index/navier-stokes-solution/?utm_source=slatesource) AI agents deployed ~10,000 autonomous agents Time to proof 88 hours (September 1 to 5, 2026) Messages exchanged between agents ~5 million Compute cost estimate Several million dollars Millennium Prize at stake $1,000,000 (Clay Mathematics Institute) Remaining Millennium problems after this 5 of 7 ## "Why would you ruin your career?" The timeline Buckmaster describes is damaging. OpenAI researchers contacted him between September 3 and 6, requesting calls. He says he only learned on September 6 that OpenAI had begun its own effort after hearing rumours he and Alpoge were close. His core allegation: OpenAI's model may have been trained on, or may have accessed, his private Codex sessions. "I asked whether the model had been trained on, or had access to, our sessions," Buckmaster said, "but received no satisfactory answer." He also says OpenAI researcher Sebastien Bubeck threatened him: "Why would you ruin your career?" and "If you don't want me to be nice, then I don't have to be nice." Bubeck denied this, saying he "came into the discussion following academic norms." A further allegation came from mathematician Andreas Thom, who says OpenAI also drew on his unpublished research without permission. [Nature: OpenAI claims huge maths breakthrough on a famed Millennium Problem (September 2026)](https://www.nature.com/articles/d41586-026-02842-5?utm_source=slatesource) ## Terence Tao's warning The most prominent voice in pure mathematics did not celebrate. Tao warned that AI companies are using unsolved mathematical problems as "marketing proof points" rather than advancing understanding. His concern is structural: AI produces verified answers without insight into why those answers work. A Lean-checked proof that no human can read or extend is a curiosity, not a contribution to the discipline. The Clay Mathematics Institute, which awards the prize, has not yet commented. The prize rules require publication in a peer-reviewed journal and two years of community scrutiny before payment. Whether Buckmaster and Alpoge receive credit as co-discoverers, and whether the credit dispute affects eligibility, remains entirely open. > The indiscriminate strip-mining of open problems for solutions may destroy the ecosystem from which the next generation of mathematical techniques, problems, and practitioners would have developed. Terence Tao, mathematician > > — Steph · 11th of September 2026 ## What actually changes The result does not immediately change applied fluid dynamics. Engineers already run numerical approximations that work. The Millennium Problem was always a question about mathematics itself: whether the governing equations are fundamentally well-behaved or contain a hidden pathology. OpenAI used an internal model described as significantly more capable than GPT-6 Astra, unavailable publicly, with 10,000 agents exchanging nearly 5 million messages over four days. The model was launched at the remaining six Millennium Problems on September 1, when OpenAI researchers heard the same rumours Buckmaster describes. What changes is the precedent. An AI system has produced a formally verified proof of a result that resisted every human attempt for two centuries. Whether it did so by synthesising known mathematics or by accessing work it should not have seen is the question that will take months to resolve. The $1 million is the smallest part of what is in dispute. [Quanta Magazine: AI Has Solved One of Math's $1 Million Millennium Prize Problems](https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/?utm_source=slatesource)