AI Aulis

AI NEWS IN ENGLISH · 22 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

21 stories · next issue tomorrow morning · How the paper is made →

← Issue 22 November 2026Paper

Buckmaster and Alpöge claim to have closed the loophole in OpenAI’s Navier–Stokes proof

AI-generated contentWritten by AI

github.com, scientificamerican.com · 2 sources · Sources published: 20 November 2026 · Some source publication dates are unknown

Tristan Buckmaster of NYU and Levent Alpöge of Anthropic have published a 79-page proof and a Lean formalisation of about 600,000 lines. The certificate has not been published, its conditions are checked outside Lean, and the corrected Lean package has not been rebuilt.

Buckmaster published the AI-assisted proof on Mastodon on 20 November, Peter Hall reports in Scientific American. The Navier–Stokes equations describe how fluids flow. When the equations blow up, the velocity of the fluid in their solution becomes infinite. In a real fluid, this cannot happen.

In September, OpenAI proved that the equations blow up. According to Scientific American, the proof formally met the conditions for the Clay Mathematics Institute’s $1 million prize. In OpenAI’s proof, however, an unrealistic, finely tuned force stirs the fluid, and many experts consider this a loophole in the problem statement. Buckmaster and Alpöge claim to have closed this loophole.

The proof comes with a Lean 4 formalisation. Lean is a proof assistant, a program that checks every step of a proof mechanically. The repository’s README says that separate files state the theorems, and Solution.lean proves them using a library of about 2,270 files and 600,000 lines. Comparator files specify which theorem statements must match and permit only three basic axioms.

Machine checking still does not cover the whole result. The formalisation contains no data, the README says. The paper gives a hash of the certificate, but the certificate itself has not been published; the paper says it will be published later. A separate computation using interval arithmetic verifies that the certificate meets the conditions in Appendix B of the paper. This computation is not part of the formalisation and is not in the repository.

The theorem statements were also corrected on 20 November, and the corrected package was not rebuilt. Lean checks proofs during the build, so the published version has not been checked with Lean.

Buckmaster said the write-up was left unfinished because, after last time, the pair did not dare to wait, Scientific American reports. Buckmaster has accused OpenAI of deliberately getting there first and has suspected the company of taking Buckmaster’s user data. OpenAI denies the allegations. The magazine reports that early information suggests OpenAI’s proof may have taken a different route from Buckmaster and Alpöge’s work.

Readers can download the Lean package from GitHub and build it themselves. The README warns that a full build needs a great deal of memory and time.

What this is based on

  1. github.com

    Publisher: github.com · source publication time unknown · retrieved 22 November 2026 at 05:11 (UTC+02:00) · discovery channel: GitHub

  2. AI just solved the ‘real’ Navier-Stokes problem

    Publisher: www.scientificamerican.com · published 20 November 2026 at 19:00 (UTC+02:00) · retrieved 22 November 2026 at 05:11 (UTC+02:00) · discovery channel: web

Previous coverage

Report an error in this story