Glossary
October 26, 2026

AI Programs Turn One of Mathematics' Longest Proofs Into a Form a Computer Can Check, and Find Mistakes in the Original

A math startup called Axiom says a team of AI programs took a famous proof that runs to about 15,000 pages, written by around 100 mathematicians over decades, and rewrote it in 24 days so that a computer can check every step. The work turned up 34 notable mistakes or gaps in the published proof, including one that needed a new argument to repair. Outside experts have not yet checked the computer version independently.

Axiom, a company in Palo Alto, California, that builds AI programs for mathematics, announced today that its programs have finished a computer-checkable version of one of the longest proofs ever written.

The result is known to mathematicians as the classification of finite simple groups, and nicknamed "the enormous theorem." It is about symmetry. Just as every molecule is built from a short list of chemical elements, every finite pattern of symmetry is built from basic pieces called simple groups. The theorem says exactly what all those basic pieces are: a handful of endless families, plus 26 oddballs that fit no family. The proof was assembled from more than 500 articles by roughly 100 mathematicians over several decades, and runs to about 15,000 pages. It is so long that very few people have ever been able to read all of it.

Axiom set dozens of AI programs to translate that proof into Lean, a computer language for mathematics. Lean accepts a proof only if every step follows from the ones before, so a finished Lean proof does not depend on anyone having read it carefully. The job took 24 days and produced about 43 million lines of Lean.

The translation exposed what Axiom calls 34 noteworthy departures from the published record. Most were small: typos, errors in tables, and assumptions that had been left out. The most serious was in one piece of the proof, where the AI programs found examples showing that a step the original argument relied on was false. Axiom says it wrote a replacement argument for that piece, drawing on ideas from the mathematician Ronald Solomon, so the overall result still holds. Solomon and another group theorist, Richard Lyons, were consulted on the work.

There are limits. A computer can confirm that every step follows, but people still have to confirm that the statements the computer was given really say what mathematicians mean; Axiom itself warns that a proof built on a hidden, unproven assumption can still pass the check, and that it had to throw out early work for exactly that reason. As of today, the claims come from the company, and no outside group has published its own review of the computer version.

Sources

  1. Axiom — Formalizing the Enormous Theorem (announcement)axiommath.ai · primary
  2. Crypto Briefing — Axiom's AI agents formalize a 15,000-page math proof in 24 dayscryptobriefing.com
Follow the timeline
Get an email when new entries are added.
© 2026 Sugarpine