AI Now Runs a Library of Math Proofs Computers Can Check

Most math lives in textbooks and papers, written in ordinary language. That's fine for humans, but a computer can't tell whether a proof is actually correct — a single hidden mistake can sit unnoticed for years. So a small but growing group of mathematicians rewrites proofs in a strict computer language called Lean (think of it as a spell-checker for logic: every step must be justified or the software rejects it). The result is math that is guaranteed correct, but only after enormous human effort.

The new project, called Lean Pool, tries to remove that bottleneck. Instead of people doing all the tedious work, AI programs — software that can take actions on its own — handle the unglamorous parts: importing existing proof collections, connecting them together, cleaning up duplicates and keeping everything organized. A single researcher, Vasily Ilin, describes it in a 52-page paper posted online, and the archive keeps growing with little human hand-holding.

So what does that actually get you? Verified math is the hidden plumbing behind a lot of things you rely on: encryption that protects your bank login, software that flies planes, chips designed to work correctly the first time. When proofs are machine-checked, those systems can be trusted rather than hoped for. If AI can maintain that foundation cheaply, verifying real-world software could become routine instead of a specialist luxury.

The catch is that this is an early, unreviewed preprint — no independent experts have checked it yet. The AI isn't inventing brilliant new mathematics; it's arranging and maintaining what humans already proved. And a machine-checked proof is only as good as the assumptions fed in. It's a promising signpost, not a finished revolution.