OpenAI says its internal model produced a mathematical flow that becomes unbounded in finite time. The company released the claim on September 8 alongside a 166-page paper and a public Lean repository. But Clay still lists the Navier–Stokes problem as active, so the announcement has not ended the dispute. On the Navier–Stokes Millennium Prize Problem Finite Time Blowup for Navier–Stokes Navier-Stokes Equation
The short answer is that OpenAI claims a major result, not that the original unforced problem has been solved. Its construction uses a smooth external force and targets alternatives C and D in the official formulation. The paper describes velocity becoming unbounded while kinetic energy remains bounded; it does not show that the unforced equations blow up. On the Navier–Stokes Millennium Prize Problem Finite Time Blowup for Navier–Stokes Navier-Stokes Equation
A proof can be checked before it is accepted
OpenAI’s Lean files matter because they encode statements and dependencies for a proof assistant to check. That is narrower than independent mathematicians agreeing that every definition, assumption and claimed implication matches the informal argument. The dossier has not independently audited either the 166-page proof or its formalization. Finite Time Blowup for Navier–Stokes Lean certificates accompanying Navier-Stokes and Euler results Navier-Stokes Equation
That distinction explains the uneasy status of the claim. A machine check can catch errors in the encoded route, while human reviewers still ask whether the encoding captures the intended theorem and whether the result answers the problem under discussion. Clay recognition and broad acceptance remain separate steps. Navier-Stokes Equation Lean certificates accompanying Navier-Stokes and Euler results
The new challenge narrows the claim, not necessarily breaks it
On September 17, Constantin, Ignatova and Vicol reported regularity under a stronger condition: the external force must be real-analytic. Their result indicates that OpenAI’s force cannot preserve the same singular behavior near the singularity under that condition. It does not, by itself, refute a construction whose stated assumption is smooth forcing. Regularity of asymptotically axisymmetric solutions to the 3D Navier-Stokes equations with analytic forcing Finite Time Blowup for Navier–Stokes
The proof has a credit problem too
OpenAI reports that about 10,000 concurrent agents worked for roughly 88 hours before further Lean verification. Tristan Buckmaster says he and Levent Alpöge used several language models to extend ideas originating with Diego Córdoba and Luis Martínez-Zoroa. Those accounts make authorship and priority part of the story, not an afterthought. On the Navier–Stokes Millennium Prize Problem Statement by Tristan Buckmaster
Buckmaster attributes the line “Why would you ruin your career?” to an OpenAI researcher during a disputed conversation. OpenAI’s published account denies accessing the specific user data and says its proof was developed independently. The dossier supplies conflicting accounts, not an established finding of misconduct. Statement by Tristan Buckmaster On the Navier–Stokes Millennium Prize Problem OpenAI pode ter feito uso não autorizado de dados para solução de Problema do Milênio
What remains to be decided
The enduring mechanism is review: a formal system can verify a precisely encoded chain, but mathematicians must still decide what that chain says and how it connects to the stated problem. For now, the strongest supported description is a claimed proof of forced alternatives C and D, accompanied by formal artifacts and facing independent scrutiny—not a recognized solution to the unforced Millennium problem. On the Navier–Stokes Millennium Prize Problem Finite Time Blowup for Navier–Stokes Lean certificates accompanying Navier-Stokes and Euler results Navier-Stokes Equation Regularity of asymptotically axisymmetric solutions to the 3D Navier-Stokes equations with analytic forcing
