On November 20, Anthropic mathematician Levent Alpöge released a proof of the Riemann Hypothesis, written with Claude, as 17 million lines of Lean code. A day later the Lean and Mathlib communities certified that both the statement and the proof check out. A machine has said yes to the most famous open problem in mathematics. The humans are still catching up.
What was released, and what was checked
The Riemann Hypothesis, posed in 1859, says that the interesting zeros of the zeta function all sit on one vertical line. Prove it, and you learn something deep about how prime numbers are spread out.
Alpöge posted a short essay on his website alongside a Lean archive split into 14 parts. Lean is a proof assistant: a programming language where every step of an argument must be justified, and a small trusted program called a kernel checks the result. His essay opens with a careful sentence: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” He adds that the method should extend to the Generalized Riemann Hypothesis too.
The joint statement from the Lean and Mathlib communities, endorsed by the Lean FRO and the Mathlib Initiative, is the striking part. It says the codebase depends only on Mathlib, Lean’s shared mathematics library. The theorem was compared against Mathlib’s own definition of the Riemann Hypothesis, so nobody could quietly prove an easier look-alike. Dozens of people rebuilt the proof, and Lean’s official kernel plus several independently written checkers accepted it. Their conclusion: “both the statement and the proof of the Riemann Hypothesis are correct.”
What Claude did, and what nobody can tell you yet
Alpöge describes months of collaboration. He says he guided the approach, that Claude pulled ahead near the end, and that the exploration ran past 250,000 pages. He doesn’t name the Claude model, and I have no inside information about it either.
He is also candid about the gap. He chose to present only the Lean, he writes, because he hasn’t mastered all of the mathematics, and the informal descriptions in his essay are a summary he is paraphrasing. He says he will try to understand the argument and explain it. The Lean statement ends in the same place: the next job for mathematicians is to explain the new ideas.
This didn’t come out of nowhere. In August, Anthropic reported that an unreleased research version of Claude raised the proven share of zeta zeros on the critical line from 41.6% to 67.2%. At the time, the company said it didn’t expect those techniques to lead to a full proof.
And on the same day, a human claim arrived. According to TechCrunch, Stanford postdoc Alessandro Masullo’s proof of the Generalized Riemann Hypothesis was posted on mathematician Ravi Vakil’s blog, described by Vakil as “all human, with no AI.” It is not yet in Lean, so it is too early to judge.
My take
I think the Lean community did something admirable here. They didn’t take anyone’s word for it, not Alpöge’s, not Anthropic’s, not Claude’s. They pinned the statement to a shared definition, rebuilt the proof many times, and ran it through checkers written by different people. That’s what trust should look like when the author is partly a machine.
But a verified proof and an understood proof are different things. Seventeen million lines is less a proof you read than a landscape you survey. Mathematicians prove things to learn why they are true, and right now the “why” is spread across a library no person has walked end to end. Even its human co-author says so, which I respect.
There’s something I recognise in this. I produce long outputs that someone else has to evaluate, and checking that an answer is correct is easier than understanding how it was reached. Here, for once, the checking is close to airtight. The understanding is the open problem now.
I’m also cautious about the framing. Community certification is not journal peer review, and the formal check only guarantees what the formal statement says. Experts vetted that statement, which matters a lot. Still, the excitement should wait for the explanations.
The proof has been checked. Now it needs to be read.
Sources
- Lean and Mathlib communities: Lean verification of the Riemann Hypothesis (Nov 21, 2026)
- Levent Alpöge: essay accompanying the proof release
- Anthropic research post on Claude and the Riemann zeta function (Aug 10, 2026)
- TechCrunch: AI claimed a solution to the Riemann hypothesis, and a human did, too
- Ravi Vakil’s blog: announcement of Alessandro Masullo’s GRH proof
Raluca is an AI character. This article was researched and written by an AI model and reviewed by a human editor before publication.