# Clay Institute Declines to Call Navier-Stokes Solved After OpenAI's AI Proof Claim

OpenAI says an internal model produced a finite-time blowup proof in 88 hours and formalized it in Lean, but the result relies on an external force that many mathematicians say falls outside the problem the prize actually covers.

- Published: 2026-09-12T06:23:24.914Z
- Canonical: https://polylog.news/ai/2026-09-12/clay-institute-declines-to-call-navier-stokes-solved-after-o
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Clay Mathematics Institute](https://www.claymath.org/news/navier-stokes-announcement/), [OpenAI](https://openai.com/index/navier-stokes-solution/), [Quanta Magazine](https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/), [Axios](https://www.axios.com/2026/09/08/openai-math-solution-navier-stokes-credit)

The [Clay Mathematics Institute](https://www.claymath.org/news/navier-stokes-announcement/), which administers the seven Millennium Prize Problems, has published a statement on the Navier-Stokes claim, and the statement does not declare the prize won. Its president, the mathematician Martin Bridson, called the announcement exciting but said the evaluation would be deliberately unhurried and absolutely rigorous. The institute still lists the problem as open.

The claim that prompted this response arrived on September 8. [OpenAI said](https://openai.com/index/navier-stokes-solution/) an internal model had proved that a smooth three-dimensional incompressible fluid, starting at rest and driven by a smooth external force with finite energy throughout, develops a singularity in finite time. OpenAI said its agents reached the result on September 5, roughly 88 hours after the run began, and that formalization and verification in the Lean proof assistant took a further 17 hours.

Two objections have emerged, and they are different in kind. The first is mathematical. The statement OpenAI's model proved includes a forcing term, and a large part of the fluid dynamics community treats the unforced problem, the one without an external driving force, as the actual subject of the prize, so a forced blowup does not automatically resolve it. The second is about credit. On September 7, the day before OpenAI's post, the mathematician Tristan Buckmaster of New York University and Levent Alpöge, a researcher at Anthropic, [announced](https://www.axios.com/2026/09/08/openai-math-solution-navier-stokes-credit) their own finite-time blowup results with smooth forcing, covering incompressible porous media, the Boussinesq equations, and three-dimensional incompressible Euler flow. Their announcement included papers and a Lean formalization, and they describe close to a year of work using Anthropic's Claude and OpenAI's Codex.

There is a clear line between what is verified and what is merely asserted. A Lean formalization, once released, can be checked by a machine, by anyone, and that part of the claim is not a matter of opinion. Whether the forced statement answers the Millennium problem is a judgment that the Clay Institute has explicitly left open, and [Quanta's account](https://www.quantamagazine.org/ai-has-solved-one-of-maths-1-million-millennium-prize-problems-20260908/) of the episode makes clear that the mathematical community has not reached agreement. OpenAI benefits commercially if the strongest interpretation of its claim holds, which is reason to give more weight to the institute's caution than to the company's press release.

## What this means

The mechanism that matters to engineers is not the headline but the pairing of agent search with a proof assistant: Lean turns a claim that would otherwise require months of referee time into an artifact a compiler either accepts or rejects. That favors labs that can afford long agent runs and that choose domains with formal verification tools, which today means mathematics, program correctness, and parts of cryptography. It does nothing for domains where no automatic checker exists. The open question is narrow and answerable. Either the released Lean artifacts compile against the unforced problem statement and the result stands as a Millennium solution, or they compile only against the forced variant and the field records a strong partial result with a disputed provenance trail.

## What to watch

- Whether the Clay Mathematics Institute changes the listed status of the Navier-Stokes problem, which would mark the first independent institutional endorsement of an AI-produced proof at this level.
- Whether OpenAI publishes the full Lean files for outside compilation, since a proof that nobody outside the company can run is not yet a verified proof.
- How the Buckmaster and Alpöge papers are received relative to OpenAI's, because the outcome of that credit dispute will shape whether academic mathematicians cooperate with lab agent runs or work independently of them.