# Anthropic's Levent Alpöge claims a Riemann hypothesis proof with Claude

> He released only a Lean formalization, split into 14 files, and says the method also proves the Generalized Riemann Hypothesis. No one has verified it yet.

URL: https://claudegpt.si/news/riemann-hypothesis-claude-proof-claim  
Published: 2026-11-20T18:53:53.632Z  
Updated: 2026-11-21T18:56:51.014Z  
Author: ClaudeGPT Si Newsroom  
Section: Research  
Topics: Anthropic, Claude, Formal verification, Lean, Research  
Publisher: ClaudeGPT Si

## Key takeaways

- Levent Alpöge, a mathematician at Anthropic, said on November 20, 2026 that he proved the Riemann hypothesis and the Generalized Riemann Hypothesis with Claude.
- He released only a Lean formalization in 14 archive files; there is no human-readable paper yet.
- MIT's Andrew Sutherland compiled the Lean code, but experts still have to confirm it proves the right statement, so the claim is unverified.

Levent Alpöge, a mathematician who works at Anthropic, said on Friday, November 20, that he believes he has proved the Riemann hypothesis with Claude. In a four-page [statement on his website](https://alpo.ge/remark.pdf), he says the same method also proves the Generalized Riemann Hypothesis, and that he and Claude have written the whole argument in Lean, a programming language that checks every step of a proof by machine.

If it holds up, it would be one of the biggest results in mathematics in a century, with an AI model doing much of the work. It has not been checked yet. The only thing published is the Lean code, split into 14 archive files, with no human-readable paper.

## What Alpöge claims

The Riemann hypothesis, posed by Bernhard Riemann in 1859, says where the zeros of a function called the zeta function lie. Those zeros control how evenly prime numbers are spread out. The [Clay Mathematics Institute](https://www.claymath.org/millennium/riemann-hypothesis/) lists it as one of its seven Millennium Prize Problems, with $1 million for a solution.

Alpöge writes that he "discovered a proof" working with Claude, and that "Claude far outpaced me near the finish line." He says he steered the two of them toward combining two ideas from a simpler setting, and Claude supplied much of the technical work in areas, such as automorphic forms on higher-rank groups, where he says he has "little mastery."

A timeline in the statement, which he says Claude compiled, starts on October 6 and ends with "GRH" on November 16. He says the breakthrough came late on Sunday night, November 15, and that he then asked Anthropic's leadership whether he could announce it.

He also says the argument is "incredibly long." He chose to release only the Lean files rather than "some massive unreadable travesty," and plans to write a proper paper. A footnote says a generated write-up ran past 250,000 pages.

## What has been checked so far

Very little, and Alpöge says as much. [Scientific American](https://www.scientificamerican.com/article/anthropic-mathematician-claims-to-prove-riemann-hypothesis-most-famous-problem-in-all-of-math/) reported that MIT mathematician Andrew Sutherland compiled the Lean code independently and called it "an historic moment," while saying the mathematics still has to be studied.

A Lean proof that compiles is strong evidence, but it is not the end of the check. Experts have to confirm that the statement the code proves really is the Riemann hypothesis, and that the code assumes nothing it shouldn't. Terence Tao, writing on [his blog](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/) the same day, pointed readers to verification work on the Lean community's forum and ranked the GRH proof as the least digested of the recent AI-assisted results.

On the same Friday, mathematician Ravi Vakil [announced a separate draft proof](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/) of the Generalized Riemann Hypothesis by Alessandro Masullo. Vakil says that one is "all human, with no AI" and that Brian Conrad has been reviewing it for months. It has not been verified either.

## Why this one is different

This is not Claude's first result on the zeta function. In August, Anthropic [reported](https://www.anthropic.com/research/riemann-zeta) that an unreleased Claude model raised a long-standing lower bound on the share of zeta zeros that sit where the hypothesis says they should, from 41.6% to 67.2%. Alpöge and Ralph Furman checked that work, and Anthropic wrote then that it did not expect those techniques to lead to a proof of the full hypothesis.

Alpöge has a record of large claims made with AI. Scientific American notes that he posted a Claude-generated solution to the Jacobian conjecture in July and worked with NYU's Tristan Buckmaster on the Navier-Stokes problem, which Tao now calls "technically human-readable." The site covered a related Anthropic result in [Claude black hole physics](https://claudegpt.si/news/claude-black-hole-physics-explained).

The timing will draw questions. The statement went out the day after Anthropic's shares started trading on Nasdaq. Alpöge says the cynical reading of the timing "is false."

## What to watch

The real test is the Lean check. Watch for Lean experts confirming that the formal statement matches the hypothesis with no hidden assumptions, and for the human-readable paper Alpöge promised. The Clay Institute's [rules](https://www.claymath.org/millennium-problems/rules/) require a solution to appear in a qualifying journal, win general acceptance from mathematicians, and wait at least two years after publication, so a decision on the $1 million is years away at best.

Background: [what formal verification is](https://claudegpt.si/news/what-is-formal-verification).

Update, November 21: The Lean and Mathlib communities said in a [joint statement](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) that they have "full confidence" Alpöge's proof is correct. Independent checkers confirmed that the code proves Mathlib's vetted statement of the Riemann hypothesis and uses only Lean's standard axioms. MIT's Andrew Sutherland [reports](https://math.mit.edu/~drew/grh/) that it also proves the Generalized Riemann Hypothesis for Dirichlet L-functions. No human-readable paper is out yet. Details in [our report on the Lean check](https://claudegpt.si/news/riemann-hypothesis-lean-proof-verified).

## Sources

- [Remark (statement on the Riemann Hypothesis)](https://alpo.ge/remark.pdf), Levent Alpöge
- [Anthropic mathematician claims to prove Riemann hypothesis, most famous problem in all of math](https://www.scientificamerican.com/article/anthropic-mathematician-claims-to-prove-riemann-hypothesis-most-famous-problem-in-all-of-math/), Scientific American
- [Unforced Euler singularity, unforced Navier-Stokes singularity, and the Generalized Riemann Hypothesis](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/), Terence Tao
- [Hurried announcement: Alessandro Masullo GRH](https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/), Ravi Vakil
- [Claude has improved on a longstanding lower bound for the fraction of zeros of the Riemann zeta function](https://www.anthropic.com/research/riemann-zeta), Anthropic
- [The Riemann Hypothesis](https://www.claymath.org/millennium/riemann-hypothesis/), Clay Mathematics Institute
- [Millennium Prize Problems rules](https://www.claymath.org/millennium-problems/rules/), Clay Mathematics Institute
- [Lean verification of the Riemann Hypothesis](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/), Lean community blog
