On September 11, 2026, the Clay Mathematics Institute (CMI) published a statement expressing hope, together with the global mathematical community, in response to the announcement that the Navier–Stokes problem "appears to have been solved." The statement follows OpenAI's September 8 release of a proposed AI-generated proof, and it is the first view offered by the institution that administers the Millennium Prize Problems. CMI says it will not rush to evaluate the result or to credit its authors, and it has not said that a prize has been decided. The public manuscript claims a mathematical breakdown of a fluid to which a smooth external force is applied, so it helps to separate what is said to have been solved from what remains to be evaluated.
A Fluid at Rest, Driven by a Smooth External Force
OpenAI's published paper, "Finite time blowup for Navier–Stokes," is a mathematics manuscript on three-dimensional incompressible fluids. The author is listed as OpenAI. A public manuscript and a formalization in Lean are available, but the materials checked on September 15 contain no journal, DOI, or record of having passed peer review. This is not a bare corporate announcement: a manuscript has been released that can be examined for the substance of its proof.
Theorem 1.1 of the manuscript claims that, for a positive viscosity, one can start a fluid with zero initial velocity, apply a smooth external force, and construct a solution whose velocity diverges in finite time. Incompressibility is the assumption that the fluid's volume does not change when it is squeezed. The force is chosen to act only within a limited range of space and time, and it is required to remain smooth through the time at which the velocity diverges.
What diverges is the fluid's velocity; the applied force is not made infinite. According to the manuscript, the solution exists and is smooth for times from 0 up to but not including 1, and as time approaches 1 the magnitude of the velocity becomes unbounded, while the total kinetic energy stays uniformly bounded. The "1" here is a time set for the mathematics; it does not mean real water reaches infinite speed after one second.
The question at issue is whether a flow satisfying certain premises can be constructed. It is not a report of a phenomenon observed in a fluid experiment, nor the result of testing a finite set of conditions in a numerical simulation. The conclusion is not judged by a sample size such as the number of subjects or experiments; the premises of the theorem and the correctness of the proof determine its scope. It also differs in nature from studies that infer causes from correlations.
Velocity Diverges, but Total Energy Stays Finite
Section 2 of the manuscript describes the construction of a vortex that draws fluid inward while also flowing along its axis. The central region where the violent motion occurs shrinks, so even as the velocity grows, the volume where the motion is concentrated gets smaller. Local velocity and the kinetic energy summed over the whole fluid are different quantities, so the divergence of the former and the boundedness of the latter can coexist.
Simply making a vortex thinner and faster cannot satisfy the condition of a smooth external force. The manuscript explains that it adds oscillations and corrections to the flow so as to cancel the parts of the terms in the equation of motion that grow large, keeping the force that finally remains smooth. This is the mechanism of the proposed proof; whether the construction is correct as a whole requires mathematical scrutiny.
The Navier–Stokes equations describe a fluid as a continuum rather than tracking it molecule by molecule. OpenAI's explanation likewise discusses velocity heading toward infinity as a breakdown of the description given by that model. The process has not been measured in everyday water or air, and it does not imply that calculations used in aircraft design or weather forecasting are all invalid.
Between an "Appears Solved" Announcement and a Prize
Comparing CMI's September 11 statement, OpenAI's public manuscript, and the prize rules shows that the assessment that the problem "appears solved," the claim of a proof with an external force, and a formal award are separate stages.
| Item checked | What the official document says | What is not settled by it |
|---|---|---|
| CMI statement of Sept. 11 | Expresses hope at the announcement that the problem appears solved; says evaluation of the result and crediting will proceed without haste | A prize decision, a recipient, or an award date |
| OpenAI's public manuscript | Claims that a construction with a smooth external force establishes CMI's options C and D | That flows with zero external force also break down |
| CMI's official problem statement | A and B ask about the existence of smooth solutions with zero external force; C and D ask about breakdown while allowing a smooth external force | An interpretation that all four must be proved |
| Millennium Prize rules | Publication in a qualifying outlet, at least two years after publication, and general acceptance by the mathematical community before CMI considers an award | That an award follows automatically two years after a corporate blog announcement |
The table matches each object of evaluation with the CMI statement, Theorem 1.1 and Corollary 10.6 of the public manuscript, page 2 of the official problem statement, and Sections 4 and 7 of the prize rules, all checked on September 15, 2026. It is not a table showing an independent audit of the proof itself.
The official problem statement by Charles L. Fefferman asks for a proof of just one of four options. C concerns breakdown in the whole of three-dimensional space, and D concerns breakdown in a periodic setting where the spatial state repeats; a construction with an external force could therefore resolve a prize problem if the claim is correct and meets the requirements. On the other hand, the conclusion cannot be extended to the case of zero external force. Satisfying the formulation of a prize problem is not the same as settling every question about fluids.
The award rules are also not simply a matter of waiting two years. Publication is normally required in a mathematical outlet with expert peer review, though CMI has discretion to relax the outlet requirement. In addition, CMI decides whether to proceed to detailed review based on whether the work has withstood rigorous scrutiny by the mathematical community. The September 8 blog announcement therefore cannot simply be treated as the starting point of the waiting period.
From a Published Formalization to a Proof People Understand
OpenAI says the group that produced the result ran roughly 10,000 AI agents in parallel, reached a solution about 88 hours after first launch, and spent a further 17 hours on formalization and verification in Lean. It also describes how humans supplied the problem options, reallocated compute, and compiled intermediate results to present back to the agents. These are the scale of work and the time reported by the company, not the results of a performance comparison with human researchers.
The public repository contains the Lean formalization and procedures for independent proof checking. The release of the code does not mean that mathematical verification by third parties has been completed. The official statement and public materials checked on September 15 contain no indication that independent third-party verification has been completed.
What CMI hopes for in its statement is that the ideas in this work will be analyzed and scrutinized, spreading new human understanding. If the manuscript's construction withstands expert examination and it becomes clear how the external force and smoothness conditions are met, the value of the result could extend beyond a verdict of right or wrong to a new way of understanding how fluids behave. The material for judging that reach will come from the mathematical discussion that works through the proof, before any prize headline.