AI Just Hit One of Math’s Hardest Fluid Problems

AI Just Hit One of Math’s Hardest Fluid Problems. The Fight Over Credit Started the Same Day

On September 8, 2026, OpenAI said an internal AI system had produced a proof that the three-dimensional Navier-Stokes equations can develop a finite-time singularity. Hours earlier, NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had posted related blow-up results for nearby fluid equations. The Clay Mathematics Institute has not announced that the $1 million Millennium Prize is awarded. The mathematics community is now reading the proofs, the Lean formalizations, and the competing timelines.

If you care about weather models, aircraft design, or what AI can actually do in research, this is the story to watch. Tell us in the comments: breakthrough, premature victory lap, or both?

What the Navier-Stokes equations actually describe

The Navier-Stokes equations model the motion of viscous fluids such as water and air. They encode conservation of mass and Newton’s second law for a continuum. Claude-Louis Navier wrote a foundational version in the 1820s. George Gabriel Stokes developed the form used today. Engineers rely on them every day. Mathematicians still do not have a complete theory of their solutions in three dimensions.

source: Wikimedia Commons

Jean Leray proved in 1934 that weak solutions exist globally. The open question is different. Start with a smooth, divergence-free velocity field that decays at infinity. Does the solution stay smooth for all future time, or can the velocity become unbounded in finite time? That second possibility is what researchers call a blow-up or a singularity.

The million-dollar question

In 2000 the Clay Mathematics Institute listed seven Millennium Prize Problems and attached a $1 million prize to each. Charles Fefferman wrote the official Navier-Stokes statement. A valid resolution is either a proof of global smoothness for admissible data, or an accepted example of breakdown. Only the Poincaré conjecture had previously been solved, by Grigori Perelman.

source: Wikimedia Commons

A singularity would not mean a glass of water explodes in your kitchen. It would mean the continuum equations themselves cease to make sense under certain idealized conditions. Simulations of everyday flows have not shown that breakdown. The prize problem is about the mathematics of the model, not a lab demonstration.

AI enters the arena

Tristan Buckmaster, a professor at NYU’s Courant Institute, already had a major mark on this field. In 2019 he shared the Clay Research Award with Vlad Vicol and Philip Isett for work on weak solutions and Onsager’s conjecture. On September 7 and 8, 2026, Buckmaster and Levent Alpöge announced finite-time blow-up results, with heavy use of large language models, for forced versions of the incompressible Euler, Boussinesq, and incompressible porous media equations. Those equations sit next to Navier-Stokes on the research map.

OpenAI then published its own claim. According to the company, an unreleased model “significantly more capable than GPT-6 Astra,” running as a coordinated group of about 10,000 agents, produced an analytical proof and a Lean formalization. The agents reached a result after roughly 88 hours. Formal verification took additional hours. OpenAI says an initially smooth fluid at rest, driven by a smooth force and keeping finite energy, can form a singularity. The described mechanism is a vortex that spirals inward and stretches, “like spaghetti.” The company says this establishes statements C and D in Fefferman’s official formulation. OpenAI also says it does not intend to claim the prize.

Watch a clear explainer of the equations and the prize problem: Navier-Stokes Equations - Numberphile

Why the credit fight matters

Multiple outlets reported a dispute over timing and method. Buckmaster said OpenAI learned of his and Alpöge’s direction and then poured compute into a similar forced-blow-up path. He raised questions about Codex logs and about proposed publication arrangements. OpenAI researchers, including Sébastien Bubeck and Mark Chen, said they did not search user data to solve the problem, that the proofs differ in important ways, and that they recognize the priority of Buckmaster and Alpöge on forced Euler. The company also wrote that it cannot fully rule out that de-identified product usage helped improve models.

That argument is not a footnote. If frontier labs can watch unpublished work through their own tools, then the norms of mathematical priority collide with the business of AI products. Readers should treat both the proof claims and the process claims as under review, not as settled history.

What is verified, and what is still open

Verified as of September 8, 2026:

• OpenAI published a writeup and said a Lean formalization exists.
• Buckmaster and Alpöge published related forced blow-up results and described AI assistance.
• Independent coverage appeared in The New York Times, Quanta Magazine, Nature, New Scientist, WIRED, and others.
• The Clay Mathematics Institute page for the problem still listed it as unsolved at the time of those reports. Acceptance of a Millennium Prize solution is a separate institutional process.

Not verified: that the Clay Institute has awarded the prize, that every mathematician accepts the argument, or that the unforced, unforced-force-free case is fully settled in the same way every reader might assume from a headline.

The result, if it holds, would be the most consequential research-level theorem yet attributed primarily to an AI system. It would also leave engineers using the same equations they used yesterday. Models that work in practice can still hide deep analytic gaps.

If this kind of story is your thing, like the post, drop a comment with the question you want answered next, and subscribe so you do not miss the peer-review aftermath.

You can write and read similar articles at the Cizeex app.

#NavierStokes #MillenniumPrize #OpenAI #ArtificialIntelligence #Mathematics

Sources

• Clay Mathematics Institute, Navier-Stokes Equation page and Fefferman official problem statement
• OpenAI, “On the Navier–Stokes Millennium Prize Problem,” September 8, 2026
• Quanta Magazine, “AI Has Solved One of Math’s $1 Million Millennium Prize Problems,” September 8, 2026
• The New York Times, “What Is Navier-Stokes, the ‘Millennium’ Math Problem OpenAI Claims to Have Solved?”
• Nature, New Scientist, WIRED, The Verge, BBC, The Guardian coverage of the same day’s announcements
• Tristan Buckmaster faculty page, Courant Institute, NYU