Back to Home

Claude Claims the Riemann Hypothesis. Peer Review Pending

🎉 Somebody has finally solved the Riemann hypothesis. Allegedly. On a Friday. In a blog post. With Claude holding the flashlight.

On November 20, Anthropic mathematician Levent Alpöge published a statement on his personal website that opened with the kind of sentence that makes entire math departments spit out their coffee: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." He went further, saying the method extends to the Generalized Riemann Hypothesis and that he expects the Extended version with more time.

For the uninitiated: the Riemann hypothesis dates back to 1859, when Bernhard Riemann guessed that the zeros of the zeta function all sit on one specific vertical line. If that sounds niche, it is not. The hypothesis governs how prime numbers are distributed, and an enormous amount of number theory is built on the assumption that it is true. The Clay Mathematics Institute put a million-dollar bounty on it. For 167 years, the best mathematicians on Earth have mostly agreed they were nowhere close.

What Actually Got Posted

There is no paper. Alpöge released the proof as Lean code, the formal verification language that checks every step mechanically, packaged in 14 compressed archive parts. He skipped the conventional write-up because the argument is, in his words, incredibly long. A footnote estimates that a generated human-readable version would run past 250,000 pages. That is not a paper, that is a municipal water system.

He was also unusually candid about the division of labor. Alpöge says he guided the conceptual move from a toy setting to the right analogue, while Claude supplied expertise he did not have, particularly on automorphic forms for higher-rank groups. The argument builds on Stepanov's function-field method, where a divisor produces multiplicity, and the hard part was reproducing that effect over the integers. Fourier-analytic ideas led the pair to repeated theta lifts, with Claude proposing a chain of a certain length for an underlying sum of the same length.

In short: the human brought the strategy, the model brought a very expensive encyclopedia and inhuman patience.

The 41-Day Speedrun

The rough timeline in Alpöge's statement puts a Stepanov determinant on October 6, iterated theta lifts and related computations on November 15, and the generalized result on November 16. From first serious move to claimed proof of the biggest open problem in mathematics: 41 days. That is faster than most home renovations, which at least come with a readable floor plan.

One caveat: Alpöge notes the timeline was supplied by Claude after a search and may need revision. The timeline for the proof is itself AI-generated. Of course it is.

What the release does not include is just as notable:

  • No independent verification yet. Nobody outside the loop has confirmed that the Lean code proves what it claims to prove.
  • No Anthropic statement. The company has not commented or attached itself to the announcement, which is a choice.
  • No traditional paper. Reviewers get 14 archives and a promise that the explanation is coming.

The Tea ☕

Alpöge says he texted Anthropic CEO Dario Amodei about the proof late on Sunday night, and the document suggests he held off publishing until after the company's IPO on Thursday. The listing raised roughly $81 billion and the stock closed its first day at $97.17, up 17% from the $83 offer price.

Proof lands on Sunday. Publication waits until after the bell on Thursday. Nothing says "independent scholarship" like scheduling your breakthrough around a lockup calendar.

This is also not Claude's first swing at the primes. In August, Anthropic reported that an unreleased research version of Claude pushed a longstanding lower bound on the proportion of zeta zeros satisfying the hypothesis from 41.6% to 67.2%, while explicitly saying it had not proved anything. That run burned 31 million output tokens across two Claude Code sessions, generated 650 ideas that did not work, and then coordinated about 60 subagents on a second attempt. Six hundred and fifty failed ideas before lunch is the most relatable number in this entire story.

And Alpöge has been here before:

  1. July 2026: he posted a Claude-generated solution to the Jacobian conjecture, which kicked off math's very strange summer.
  2. September 2026: he was at the center of the Navier-Stokes spat, when NYU mathematician Tristan Buckmaster implied OpenAI had taken his and Alpöge's work.
  3. November 2026: the Riemann hypothesis, released as a zip file.

The uncomfortable modern part: Lean might settle this faster than any referee ever could, because a compiler does not care about reputations, conference deadlines, or who texted whom on Sunday. If the code checks, it checks. Scientific American notes that verification should not take long.

But a machine-checked proof with no readable argument is a strange artifact. Mathematics is supposed to be a body of explanations, not a binary that returns true. If the only entity that fully understands the proof is the model that helped write it, the field has a new problem to solve right after it finishes celebrating this one.

MIT mathematician Andrew Sutherland once put the stakes about as well as anyone: "An AI that can prove the Riemann hypothesis is not one I'd want to meet. If that happens, mathematicians having jobs will be the least of our problems."

Alpöge, for his part, is not gloating. His statement is almost elegiac. "It is natural to feel that this is an end, in my case a loss of a principal guiding light," he wrote, "but what I saw during the course of progress so amazed me that I believe it is one dawn across science of something far greater."

Maybe Claude really did it. Maybe the Lean compiler is about to have the worst weekend of its life. Either way, the confetti stays in the box until a human who does not collect an Anthropic paycheck says the words "it checks." Until then, this is the most impressive "trust me, it compiles" in the history of mathematics.

Your friendly neighborhood peer reviewer, unpaid and unqualified.

Comments

No comments yet. Be the first to share your thoughts!