Skip to content
See the World Through ScienceA project of ALLATRA

Google Says Its AI Agents Rewrote a Landmark Math Proof as Code a Computer Can Check

AI & Technology

Republish this story

Our work is licensed under Creative Commons BY-NC 4.0. You may republish this piece for free — with credit to ALLATRA Media and a link to the original, unedited beyond length trims, and not for commercial use.

Read the full license

A woman writing dense handwritten mathematical formulas in colored marker across a large classroom whiteboard.
A proof worked out by hand, a line at a time, photographed May 19, 2021 (illustrative). Google says its agents instead produced about 1.8 million lines of Lean code that a computer checks step by step."Natalia Krizhanovskaya writes mathematical formulas on the blackboard 2021" by English: Andrew Krizhanovsky Русский: Андрей Крижановский, via Wikimedia Commons, Public domain

Google Research said on Nov. 13, 2026, that teams of its AI agents produced a complete, machine-checked version of the Weil conjectures, the landmark results on counting solutions of polynomial equations in finite number systems. The proof itself dates to 1974, when Pierre Deligne completed it. What the agents produced is that existing proof rewritten in Lean 4, a language that turns a proof into code a computer checks line by line.

In its announcement, Google gives its reason: checking a computer-assisted proof by hand is heavy work for peer reviewers, and small flaws can hide inside long calculations. Once mathematicians have confirmed that the formal statement matches the original problem, the code compiles only if every logical step holds.

The formalization runs to about 1.8 million lines of Lean, produced across 28,800 separate agent calls, and Google describes the run as fully autonomous. The company does not say whether that code has been released. The same framework also produced what Google calls the first formal version of a recent proof of Hilbert's tenth problem over general rings of integers. All of these figures are Google's own account of its own system, and no outside group has reproduced the runs.

One result in the post is new mathematics, separate from the formal rewrites described above. Google says a team that included researchers from the Chinese University of Hong Kong, the University of Southern California and Carnegie Mellon University proved the Courtade-Kumar conjecture in full generality. That question in information theory had been open since 2013. The post splits the credit: the models found counterexamples to a generalized version and developed most of the new bounds, and the academic team proposed the candidate function and the key bound they built on.

Google said the campaigns ran on its Gemini 4 models, among them Gemini 4 Carbon. The agent tools are now available to outside developers in Google's Antigravity environment, and the company has opened applications for 10,000 academic researchers to get six months of free access to its highest usage tier.

Sources

By Olga SchmidtEditor-in-Chief, Writer

Spot an error?

Spot an error?

Report an error

Spotted a mistake on this page? Tell us what's wrong and our editors will take a look.

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We correct mistakes openly. Select any text to flag it. Fixes are logged under our Corrections Policy.

Report an error

Reporting on

Google Says Its AI Agents Rewrote a Landmark Math Proof as Code a Computer Can Check

What kind of problem?

Only if you'd like us to be able to follow up. We won't use it for anything else.

We read every report. Corrections are logged publicly.