Riemann Hypothesis: Claude at the Heart of an Unverified Proof

Mathematician Levent Alpöge claims to have found a proof of the Riemann Hypothesis with Claude. His demonstration, published in Lean, has not yet been independently verified and cannot claim the million-dollar prize until academic validation.

Levent Alpöge, a mathematician at Anthropic, claims to have proven the Riemann Hypothesis with the help of Claude, his employer's AI assistant. The announcement, posted on November 20, targets one of the seven Millennium Prize Problems, posed in 1859 and related to the distribution of prime numbers. At this stage, the proof has not been independently verified, leaving the matter at the most delicate stage of mathematics: where the code speaks, but the community does not yet.

Une annonce qui bouscule un vieux problème mathématique

The Riemann Hypothesis has been central to number theory for 167 years. It concerns how prime numbers are distributed, a very abstract subject that nevertheless underpins much of modern mathematics. If Alpöge's proof holds, it would not only solve a famous conjecture; it would also show that an AI model can contribute to a major breakthrough in a discipline known for its caution.

The researcher says he goes even further by extending his method to the generalized Riemann Hypothesis. In his text, he credits Claude with the decisive final steps of the reasoning. The passage quoted by The Verge is explicit: "I believe I have, working with Claude, discovered a proof1 of the Riemann Hypothesis."

Pourquoi le format Lean change la donne

The demonstration was not published as a traditional scientific article. Alpöge posted fourteen code archives written in Lean, a formal proof language that allows a computer to verify each step of a mathematical argument. This is a pragmatic choice, but also a severe constraint: without human review and independent validation, a proof can remain technically elegant yet scientifically fragile.

According to his note, the mathematician gave up on releasing the version written with Claude, which he describes as a document of over 250,000 pages, too large to be usable. He promises a readable presentation later. Meanwhile, experts will have to verify two very concrete points: whether the hypothesis has indeed been translated into Lean, and whether the argument withstands scrutiny. Two steps. And not the easiest.

La récompense d’un million reste hors de portée immédiate

Even if confirmed, the prize would not be awarded immediately. The Clay Institute, which funds the seven Millennium Problems, offers one million dollars per problem, about 910,000 euros, but only after publication in a recognized journal and then a period of at least two years during which the solution must be widely accepted by the mathematical community. In other words, the inbox may be excited, but not the accountant.

The timing also draws attention. The Verge notes that the announcement came just after Anthropic's IPO, on November 19, 2026, according to its publication, and highlights that Alpöge said he wrote to Dario Amodei, the company's CEO, "late Sunday night" before publishing. The mathematician himself considers a cynical reading of the timing "obvious," while rejecting it.

Ce que cette affaire dit de l’IA en mathématiques

This announcement is part of a broader sequence where AI is advancing rapidly on mathematical problems. The Verge recalls that Anthropic had already worked on the Riemann Hypothesis with Claude, while stating in August that it did not expect the techniques used to be sufficient to prove it. OpenAI, for its part, claimed in September progress on the Navier-Stokes problem, without formal recognition at this stage. Mathematicians now have to sort announcements, proofs, and demonstrations that truly hold up.

For Anthropic, the stakes go far beyond the prestige of a hypothesis. If Claude indeed served as a decisive tool in a problem of this magnitude, it would reinforce the idea that AI models no longer merely help write text or routine code, but can also intervene in fields where formal rigor matters more than inspiration. The uncomfortable question remains: until independent verification occurs, the proof exists mainly on its author's site.

By the numbers

  • 1859 — year the Riemann Hypothesis was formulated.
  • 7 problems — the Clay Institute's Millennium Problems.
  • 1 million dollars — reward promised per problem.
  • 2 years — minimum review period before awarding.
  • 14 archives — number of code files published by Alpöge.

To read