OpenAI Claims a Navier–Stokes Proof. A Fight Over Priority Is Already Underway

OpenAI has made the kind of claim mathematicians spend careers trying to establish. The company says an internal system of agents, using a model it describes as significantly more capable than GPT-6 Astra, has proved that a smooth solution to the three-dimensional Navier–Stokes equations can develop a singularity in finite time. It has released a written proof and a formalization in Lean. That is a claim about a proof, not yet a verdict on one.

The equations describe the motion of viscous fluids and gases. The Millennium Prize question asks whether their three-dimensional solutions must remain smooth or whether a breakdown can occur. OpenAI says its construction starts from an initially smooth state of rest and reaches a singularity. The precise assumptions and every step of that construction now need scrutiny by mathematicians outside the company.

That scrutiny has been complicated by an immediate dispute over priority. Mathematicians Tristan Buckmaster and Levent Alpöge had been working with language models, including Codex, on related questions about singularities. Buckmaster says information about their progress reached OpenAI and that, when they contacted the company, they were told its internal system had already produced a full proof of the central problem.

OpenAI disputes the account. Sébastien Bubeck, who leads its mathematics work, called the allegations “false and inflammatory.” The company says it did not train its models on Buckmaster and Alpöge’s materials or use their prompts, models, or proofs. Neither account, on its own, establishes what information passed between the researchers and the company. The dispute concerns both the route to a possible result and the evidence needed to assign credit.

There is an important distinction between the work in dispute. Buckmaster and Alpöge’s publicly described preprints concern singularities in models including the incompressible porous medium equation, the two-dimensional Boussinesq system, and the three-dimensional incompressible Euler equations. Those are related problems, not a published solution by the pair to the Navier–Stokes Millennium Prize question. OpenAI’s claim addresses that central question directly. Similar research interests do not, by themselves, establish a shared proof or an act of borrowing.

The prize raises the stakes without settling anything. Navier–Stokes is one of the Clay Mathematics Institute’s Millennium Prize Problems, with $1 million allocated to a solution. Clay still lists it as unsolved. Under its rules, the institute does not accept direct submissions: before a proposed solution can be considered, it must appear in a qualifying outlet, at least two years must pass, and the result must gain general acceptance among mathematicians.

A proof announcement is not a solved Millennium Prize Problem.

A Lean formalization gives specialists another object to inspect, not a shortcut around that process. They will need to check the mathematical argument, the technical conditions of the claimed singularity, and whether the formalization establishes what the public announcement says it does. For a problem of this scale, publication and independent examination matter more than the speed of an announcement.

The priority dispute may prove harder to resolve than a flaw in a proof. It asks what the two teams knew, when they knew it, and how to describe contributions made with language models and agentic systems. For now, OpenAI has announced an extraordinary proposed result, Buckmaster and Alpöge have challenged the account of its origins, and neither the mathematics nor the competing claims of priority has received a final verdict. It is also a test of the chain of custody a major AI breakthrough may need.