AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Levent Alpöge, a mathematician who works at Anthropic, says he believes he found proofs of the Riemann Hypothesis and the stronger Generalized Riemann Hypothesis with Claude. The claim is not an accepted proof: a large Lean code archive has been released, an independent compile was reported, and a further check and human review remain outstanding.

Mathematician Levent Alpöge, who works at Anthropic, posted a note on November 20 saying he believes he and Claude proved the Riemann Hypothesis and its stronger counterpart, the Generalized Riemann Hypothesis. He also released a large archive of Lean code intended to formalize the argument, but the claim has not been accepted by mathematicians, and important checks are still pending.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

Alpöge’s four-page note, posted on his personal website, says he “believe[s]” he discovered a proof while working with Claude. He says the method extends to the Generalized Riemann Hypothesis, which concerns a broader family of related functions. The note offers a sketch and a timeline; it does not include a readable proof. Instead, it links to a Lean archive divided into 14 parts. Lean is software that checks whether a formalized argument follows from its stated assumptions.

The archive’s scale is unusually large. Computer scientist Scott Aaronson put it at 17.4 million lines of Lean code, according to the source material. Alpöge says a human-readable version would exceed 250,000 pages and that he chose not to publish it. Terence Tao described the work as an “(enormous) Lean-only proof.” These descriptions indicate the practical challenge: a computer may check formal steps, while mathematicians still need to determine whether the formalization proves the intended result and how the argument works.

Some early technical checks have been reported, but they do not settle the claim. MIT mathematician Andrew Sutherland told Scientific American that the code compiled independently “with only the three standard axioms.” Kevin Buzzard wrote that a “comparator run” was still awaited. The source describes that as a check on whether the code proves the intended statement without relying on an improper shortcut. A successful compile is evidence about the code’s formal consistency under a checking system; by itself, it does not establish that the result has passed mathematical scrutiny.

At a glance
reportWhen: Claim posted November 20, 2026; status…
The developmentA four-page note and a 14-part Lean archive posted by Levent Alpöge on November 20 set out his claim that he and Claude proved the Riemann Hypothesis.
Bar chart titled 20 steps in six weeks, eight of them on one day. Steps per date in the timeline of Alpöge's note: October 6, 4; October 8, 1; October 9, 1; October 14, 2; October 19, 2; October 20, 1; November 15, 8; November 16, 1, the last step, which reads GRH, highlighted
The 20-step timeline in Alpöge's note, which he says Claude compiled, by date. Source: Levent Alpöge, remark.pdf, pages 3 and 4, November 20, 2026. Chart: Thorsten Meyer AI.

Why a Proof Would Change Number Theory

The Riemann Hypothesis is one of mathematics’ most prominent unresolved problems. It concerns the locations of the nontrivial zeros of the zeta function and is closely connected to the distribution of prime numbers. A proof would settle a question that has shaped number theory since Bernhard Riemann’s 1859 work and would have consequences for results that depend on estimates about primes.

The Clay Mathematics Institute lists the hypothesis among its Millennium Prize Problems, each associated with a $1 million prize. That prize is not triggered by an online claim or a computer archive alone. Clay’s rules, as described in the source material, call for publication in a refereed journal of worldwide repute, at least two years of scrutiny after publication, and general acceptance by the mathematical community. The current claim has not reached those stages.

The episode also puts a specific question about AI-assisted research before mathematicians: whether a formal proof produced with substantial machine assistance can be checked, understood and evaluated by people. Alpöge says Claude did much of the technical work near the end, while he steered the strategy. That account is his description of the collaboration, not an independent assessment of the model’s contribution.

Products Worth Considering

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

Anker SOLIX C1000 Gen 2 Portable Power Station with 200W Solar Panel

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

LG UltraGear 34-Inch Curved QHD Gaming Monitor, 160Hz, FreeSync Premium

As an affiliate, we earn on qualifying purchases.

From Riemann’s Question to Lean

In 1859, Bernhard Riemann linked the zeta function to patterns in the prime numbers and proposed that all of its nontrivial zeros lie on a particular vertical line in the complex plane, where the real part is one-half. The hypothesis remains unproved in the account provided here. Its importance comes in part from the way the zeros constrain how regularly primes are distributed.

Formal proof systems such as Lean allow arguments to be encoded as steps a small checking program can verify. Such verification can help detect gaps in a formalized derivation, but the formal statement, definitions and assumptions still matter. In this case, the source reports a note, the Lean archive, an independent compile, and an outstanding comparator run. It does not report journal publication or broad expert acceptance.

The note’s author is an Anthropic employee, and the claim concerns work with Anthropic’s AI. The source says Anthropic had not announced the result on its news page as of November 21 and that Nature reported it had not immediately responded to a request for comment. No Claude model version is identified in the note, according to the material provided. A second, rival claim by a Stanford mathematician also appeared within hours, but the source excerpt supplies no details sufficient to assess that claim.

“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis”

— Levent Alpöge, in his November 20 note

Key Checks Still Await Review

The source material does not establish that the Lean archive has passed the pending comparator run. That check matters because a formal proof is only meaningful if its encoded statement matches the mathematical claim and its assumptions are appropriate. The reported independent compile is an early result, but the available account does not provide the comparator outcome or a full technical analysis of the code.

It is also unclear how experts will assess the proof’s substance. Alpöge says he has “little mastery” of one of the main areas used in the argument, and the note does not provide a conventional readable proof. The source reports sharply different early reactions: Tao called it an “(enormous) Lean-only proof,” while Andrew Granville called the released note “garbled.” Neither description resolves whether the formal argument is correct.

No peer-reviewed publication, general mathematical acceptance, or Clay prize decision is reported. The note does not identify which Claude model was used. Anthropic had not publicly announced the result in the source’s check on November 21, and Nature reported that the company had not immediately responded to its request for comment. These points may change as further information and review become available.

Independent Review Must Continue

The next reported technical milestone is the comparator run that Buzzard said was still awaited. If that check is completed, its result would help establish whether the formal code proves the intended statement under the stated assumptions. Mathematicians would then need time to inspect the formalization, test the definitions and assess the argument’s relationship to the original hypotheses.

For the claim to follow the path required for Clay’s prize, it would need publication in a refereed journal of worldwide repute, at least two years of scrutiny and general acceptance by mathematicians, according to the rules cited in the source. No timetable for those steps is given. Until they occur, the accurate status is that Alpöge has made a striking claim and released code for checking; whether it constitutes a proof remains unresolved.

Key Questions

Has the Riemann Hypothesis been proved?

Not on the evidence reported here. Alpöge says he believes he and Claude found a proof, but mathematicians have not accepted it and review remains incomplete.

What did Alpöge release?

He posted a four-page note with a sketch and linked to a Lean archive in 14 parts. The source says the archive contains an estimated 17.4 million lines of code.

What is the pending comparator run?

It is a check described in the source as testing whether the formal code proves the intended statement without an improper shortcut. Kevin Buzzard said the run was still awaited.

Did Claude prove the hypothesis on its own?

Alpöge describes a collaboration: he says he steered the strategy and that Claude outpaced him near the end. The source does not independently establish how much of the work the model performed or identify the Claude model version.

Would this claim qualify for the Clay prize now?

No prize decision is reported. The Clay rules cited in the source require publication in a refereed journal of worldwide repute, at least two years of scrutiny and general acceptance by mathematicians.

Source: Thorsten Meyer AI

You May Also Like

Claude: Elevated Errors Across All Models

Recent reports indicate elevated error rates across all Claude AI models, raising concerns about reliability and performance.

The Machine Economy — Capital-Heavy, Human-Light, Trading With Itself

Analysis of the emerging machine economy where AI-driven firms operate with minimal human labor, trading primarily with each other and reshaping the global economic landscape.

The Memento Constraint: Why Continual Learning Is the Trillion-Dollar Bottleneck Nobody Is Pricing

Exploring how the inability of current AI models to learn continually limits enterprise AI and the potential economic impact of solving this challenge.

Expanding Our Support For Scientists

A new program aims to significantly increase resources and funding for researchers across multiple disciplines, enhancing scientific progress globally.