An AI Swarm Just Checked the Longest Proof in Math

The proof nobody could check alone

Some work is too big for any one expert to verify. The Rundown AI reported that AI math startup Axiom completed a fully machine checked version of the Classification of Finite Simple Groups, described as the longest proof in mathematics.

The theorem is a catalog of the basic building pieces of symmetry, the parts every symmetry pattern in math breaks down into. More than 100 mathematicians produced the original across roughly 15,000 pages and more than 500 papers, and Axiom says no single expert has ever read the whole thing cover to cover. For decades the result stood on trust, because a full check was beyond human scale.

How the swarm did it in 24 days

Axiom broke the mountain into 108 milestones. Then dozens of AI agents, steered by the Axiom team, rebuilt the proof as about 43 million lines of Lean, a coding language built so a computer can verify every logical step instead of taking any step on faith.

The whole effort took 24 days. That pace is the headline, but the method is the story: split an impossible job into checkable pieces, let agents draft at machine speed, and let a strict verifier accept or reject every piece before it joins the whole.

The catch that proves the point

The theorem stands. But the check was not a rubber stamp. During the rebuild, the AI caught a false claim inside one key step of the original proof, a flaw that had survived decades of human scholarship, and patched it with a new proof based on 1976 work by mathematician Ronald Solomon.

That single catch justifies the entire exercise. A verification system that only agrees is decoration. This one found a real error in work the field had trusted for years, and fixed it with a proof the machine could certify.

Why this matters beyond math

Every field has its 15,000 page problem: audits, archives, codebases, contracts, and catalogs too large for one careful person to recheck. The bottleneck was never intelligence alone. It was verified attention at scale.

The Axiom pattern supplies exactly that. Agents provide the attention, the verifier provides the discipline, and humans provide the steering and the judgment about what deserves checking in the first place. That division of labor travels well beyond mathematics.

• Split oversized work into milestones a checker can accept or reject
• Pair fast agent drafting with a strict verification step, never drafting alone
• Keep humans on steering and judgment, where taste and priorities live
• Treat a caught error as proof the system works, not as an embarrassment

Verification is the product

Generation is getting cheap everywhere. What stays scarce is work a buyer can trust without rechecking it themselves. Axiom just demonstrated that trust can be manufactured with the right structure.

Build the checker into your workflow now, and your output carries a guarantee your competitors cannot match with speed alone.