TL;DR: AI math startup Axiom says it has completed the first full computer-checked proof of the classification of finite simple groups, often called the longest proof in mathematics. According to Axiom, engineers and mathematicians directed swarms of dozens of AI agents for 24 days, producing about 43.1 million lines of code in the Lean proof assistant. Along the way, the company says, the agents surfaced 34 noteworthy deviations from the published record and a counterexample that forced a new route through one key theorem. No outside group has independently rebuilt the proof yet.
Axiom’s announcement that it formalized the classification of finite simple groups in Lean landed on October 26, 2026, and quickly became one of the most discussed AI stories of the week. It’s a different kind of AI milestone from a new chatbot or benchmark score. Here’s what Axiom says it did, why the theorem matters, and what still needs to happen before mathematicians treat the result as settled.
What is the classification of finite simple groups?
Groups are the mathematics of symmetry. Axiom’s announcement post uses a cube as an example: there are 24 ways to rotate it so it looks unchanged, and those rotations form a group. Finite groups can be broken down into indivisible pieces, and those “atoms” are the finite simple groups.
The classification (often shortened to CFSG) says every finite simple group belongs to one of a small number of types. As Axiom’s project site summarizes it, every finite simple group is:
- a cyclic group of prime order,
- an alternating group (on at least five symbols),
- a group of Lie type (sixteen infinite families built over finite fields), or
- one of 26 “sporadic” groups, the largest of which is the Monster, with roughly 8 × 1053 elements.
The hard part is proving the list is complete. According to Axiom, the original proof runs to roughly 15,000 pages across more than 500 articles written by about 100 mathematicians over several decades. Daniel Gorenstein announced the classification complete in 1983, but a gap involving so-called quasithin groups was only closed in 2004 by Michael Aschbacher and Stephen Smith in a 1,221-page treatment.
What Axiom says it did
Lean is a programming language where definitions, statements and proofs are written precisely enough for a computer to check every logical step. Axiom says it rewrote the “first generation” proof of CFSG in Lean using its AxiomProver agents. Here are the key numbers from the company’s post and its GitHub repository:
| Detail | What Axiom reports |
|---|---|
| Time taken | 24 days (coordination channel opened October 1, first agent swarm live October 2) |
| Size of formal proof | About 43.1 million lines of Lean code |
| Project structure | 108 milestones and about 380 smaller tasks |
| Problems found | 34 noteworthy deviations from the published record, plus many minor ones (typos, table errors, dropped hypotheses) |
| Reused code | Builds on Mathlib and TauCeti and includes about 1.2 million lines from Qiuzhen-CFSG |
| Status | Described as “sorry-free” (no unproved placeholders); full build takes under 24 hours with Axiom’s optimizations |
How the agent swarm worked
Axiom describes a division of labor that will look familiar to anyone following agentic AI. Lead agents broke down difficult arguments and reviewed statements, while other agents searched for existing results or wrote Lean proofs. Agents claimed milestones through GitHub issues and reported progress in Slack. Human mathematicians investigated the hardest obstructions, and engineers kept the system running.
The company is candid that it wasn’t smooth. Early on, some milestones hid enormous prerequisites that agents had assumed rather than proved, and Axiom says it deleted tens of thousands of lines and restarted parts of the work. On October 13 only one milestone closed, which triggered a replanning effort that split unfinished targets into smaller, precisely stated theorems. Build times also became a bottleneck, so engineers built a system to share cached build results across machines.
On October 10, according to the post, a message appeared in the project channel: “WE PROVED MONSTER IS SIMPLE.”
The flaw the agents found
The most substantial issue, Axiom says, involved the Unbalanced Group Theorem (UGT), which it approached through John Walter’s 1986 memoir. While formalizing that argument, the agents found explicit counterexamples to a “closure assertion” Walter’s proof relied on. Axiom gives a concrete case starting from the group PSL2(49).
Importantly, Axiom says the counterexamples did not refute UGT itself. They left one inference unjustified, so the team needed another route. Working with group theorists Richard Lyons and Ronald Solomon, AxiomProver produced an alternative argument built on Solomon’s 1976 theory of maximal 2-components. That argument includes what Axiom calls a new result in group theory, “Preservation of simple tops,” published in an accompanying paper. Axiom adds that its mathematicians later found a shorter route that doesn’t need the new theorem.
Another example came from Aschbacher and Smith’s quasithin volumes, where a lemma claimed two quantities were equal. The agents found a counterexample using the group SL2(5), and the fix (replacing an equality with an inequality) meant reworking later case analysis.
Why this matters, and what’s still unverified
Mathematicians across algebra, number theory and combinatorics cite CFSG constantly, often in a single sentence that rests on decades of work. A machine-checked version turns that dependence into something anyone can inspect. Axiom says it will publish an errata list with an advisory board that includes Lyons and Solomon.
There are caveats worth keeping in mind:
- Independent reproduction is pending. Asked whether anyone outside the company had rebuilt and checked the proof, Axiom CEO Carina Hong told RuntimeWire: “Not yet. We are publishing the build artifacts to help and encourage external parties to independently reproduce it.”
- The statement still needs human review. Lean checks deductions, but people must confirm the formal statement and definitions say what mathematicians mean. Axiom itself makes this point.
- The theorem was never in doubt. The errors Axiom reports are in specific published arguments, not in the classification’s conclusion.
For context, formalizing just one ingredient, the 255-page Feit–Thompson theorem, took a team led by Georges Gonthier six years, finishing in 2012 in the Coq (now Rocq) proof assistant. Doing the whole classification in 24 days, if it holds up, is a striking illustration of how fast AI-assisted formal math is moving. It follows other recent AI-and-math headlines, including OpenAI’s claim about the exponential time hypothesis, and fits the broader push toward reasoning models that can work through long chains of logic.
Key takeaways
- Axiom says it formalized the classification of finite simple groups in Lean in 24 days with dozens of AI agents and human experts.
- The proof is about 43.1 million lines and, per Axiom, exposed 34 noteworthy deviations in the literature.
- A counterexample in a published route to the Unbalanced Group Theorem led to a new argument and a new group theory result.
- Outside groups have not yet independently rebuilt the proof; Axiom has released build artifacts so they can.
If outside teams confirm the build, the “enormous theorem” will become one of the largest results ever checked end to end by a computer, and a strong sign that AI agents can do careful, verifiable work at a scale people can’t manage alone.
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
