Ta strona nie jest jeszcze przetłumaczona, więc wyświetla się po angielsku.

Has the Riemann Hypothesis been solved? The Claude claim

Convika Team13 min read

  • math
  • claude
  • ai-news
On this page

As of November 21, 2026, 01:50 UTC (10:50 JST). Alpöge's site, the independent check's repository, Scientific American's report, Terence Tao's post, and the Clay Mathematics Institute's page were read between 01:46 and 01:55 UTC that day. This page follows the check and changes when its result is in.

On November 20, 2026, Levent Alpöge, a mathematician at Anthropic, wrote: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." He published the proof only as a Lean formalization, a form a computer checks line by line. An independent check has passed its first gates: the source matches the published archive, and it contains no sorry and no axiom. Andrew Sutherland, a mathematician at MIT, has compiled the proof in Lean on his own and calls it "an historic moment." The full machine check, Comparator, is still building, with no result yet. The Clay Mathematics Institute's page lists the Riemann Hypothesis as "Unsolved" (all as of November 21, 2026, 01:50 UTC, 10:50 JST).

Below: what was announced, what a Lean proof is and what must check out, what mathematicians say, the Riemann Hypothesis in short, what happens if the proof holds, the past claims that failed, and who gets the $1 million prize.

What was announced

Alpöge posted a 4-page note on his site, alpo.ge, on November 20, 2026. Its first paragraph:

  • The claim: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis."
  • The scope: "In fact, the method easily extends much further, resolving the so-called Generalized Riemann Hypothesis, and I expect, with more time, the Extended Riemann Hypothesis too." So the note claims RH and the Generalized Riemann Hypothesis (GRH), and states the Extended Riemann Hypothesis as an expectation.
  • The formal proof: "Claude and I have also formalized the arguments without assumptions."

Claude's role. Alpöge writes that he set the direction and Claude took the last stretch: "this time Claude far outpaced me near the finish line." He gives one idea as Claude's. The argument descends from Stepanov's method, which proved the analogue of the Riemann Hypothesis over function fields and uses "multiplicity." Over the integers, multiplicity is hard to find. For a sum of length X, Claude suggested a chain of theta lifts of length √X, each lift standing in for one step of multiplicity. The note's timeline, which Claude compiled, runs from the "Stepanov determinant" on October 6, 2026, to iterated theta lifts on November 15 and "GRH" on November 16.

Lean only, a paper to come. "I decided to only present Lean," Alpöge writes, rather than post "some massive unreadable travesty." A footnote says his drafts "hit 250,000+ pages." The archive is 14 parts, lean.tar.gz.part01 to part14. He intends "to speak about it, to write it properly," and expects the argument to be "massively simplified, likely immediately."

He writes that Anthropic's leadership and "the few trusted mathematicians I consulted" shared his view on how to announce. As of November 21, 2026, 01:50 UTC, Anthropic and the Clay Mathematics Institute had issued no statement, and alpo.ge showed no new paper.

Who is Levent Alpöge? A number theorist whom Scientific American describes as "a mathematician and employee at artificial intelligence company Anthropic." In August 2026, he and Ralph Furman posted a paper showing that more than two thirds of the zeta function's nontrivial zeros are simple and lie on the critical line, with the comment "Proof discovered autonomously by Claude (Anthropic)," formally verified in Lean 4.

What a Lean proof is, and what must check out

Lean is a proof assistant: a program that accepts a proof only when every step follows from the rules of logic and the definitions. Mathlib is its shared library of mathematics, and it already defines the Riemann Hypothesis as RiemannHypothesis. A Lean proof is accepted on 4 points:

QuestionWhat checks itStatus, November 21, 2026, 01:50 UTC
Is the published source the one being checked?Byte comparison with the published archivePassed
Does it state the real theorem?A challenge file that imports only Mathlib and states the theorems; riemannHypothesis : RiemannHypothesis uses Mathlib's own definitionThe challenge file compiles: passed. Comparator matches the statements: building, no result yet
Does it assume anything extra?A scan for sorry, axiom, native_decide, unsafe, implemented_by, opaque; Comparator's axiom checkThe scan passed. Comparator's axiom check: building, no result yet
Is every step valid?Kernel replay in Lean's kernel and in nanoda, a second, independent checkerBuilding, no result yet

The check is a third party's, at github.com/Deicyde/riemann-hypothesis. It also found that all 9 dependency pins are official upstream commits. Comparator compiles all 175,187 files itself, which takes hours and about 100 GB of disk. As of November 21, 01:50 UTC, the results section of the check's NOTES.md reads "The build is in progress." Its last commit was at 20:43 UTC on November 20, and its automated checks last passed at 20:47 UTC. Three issues are open: a smaller version for faster builds (#3), fewer imports (#4), and an import-cone check, which finds no module outside the cone of the main file (#2).

The first outside check by a mathematician. Scientific American reports that Andrew Sutherland of MIT "independently compiled Alpöge's result in Lean." He said: "This is indeed an historic moment." The full Comparator verdict is still to come.

GRH here means Dirichlet L-functions: the formal statement is that L(χ, s) ≠ 0 when the real part of s is above 1/2, for every modulus and every character χ.

What a pass trusts. The README lists it: the Lean kernel and nanoda, Mathlib's definitions, Mathlib's official build cache, and the sandbox the check runs in. So the question left for people is the one Kit Yates put on November 20: "Experts will need to check that the theorem proved in Lean really is the Riemann hypothesis," and "that the definitions faithfully capture the standard mathematical objects."

What mathematicians say

Terence Tao wrote on November 20 that Alpöge "has generated an (enormous) Lean-only proof" of the Generalized Riemann Hypothesis. From a very first look, he writes, the strategy appears to use the full power of the Langlands program and arithmetic geometry, and he expects digesting it to take time. He points readers to the Lean Zulip thread to follow the verification. In Scientific American, Hector Pasten of the Pontifical Catholic University of Chile said that if the proof is correct and understood, it "completes a dictionary" between branches of math. Carlo Pagano of Concordia University, a visiting researcher at Google DeepMind, said its ideas could "open up brand new avenues" in number theory. Sutherland added: "There is much work to be done to understand the mathematical details of the proof."

What the Riemann Hypothesis is, in short

In 1859, Bernhard Riemann's paper "On the Number of Primes Less Than a Given Magnitude" tied the primes to a function, the zeta function ζ(s). It is zero at −2, −4, −6 and so on, the trivial zeros. Its other zeros, the nontrivial ones, lie in a strip between real part 0 and real part 1. The Riemann Hypothesis says every nontrivial zero has real part exactly 1/2: they all sit on one line, the critical line.

Clay's page puts it as "all the 'non-obvious' zeros of the zeta function are complex numbers with real part 1/2," and notes it "has been checked for the first 10,000,000,000,000 solutions." Platt and Trudgian checked it up to a height of about 3·10¹². David Hilbert listed it in 1900 as part of his 8th problem. The plain-English version, with pictures of the zeros, is in What is the Riemann Hypothesis?.

What happens if it holds

  • Primes are as regular as they can be. Helge von Koch showed in 1901 that RH is equivalent to the best possible error in the prime number theorem: the count of primes up to x differs from Li(x) by at most about √x·log x.
  • Gaps between primes get a bound. Assuming RH, Harald Cramér proved that the gap after a prime p is O(√p·log p).
  • "Assuming RH" theorems become theorems. Many results in number theory are proved "assuming RH" or "assuming GRH." They become unconditional. One example: in 1976 Gary Miller showed that under GRH his primality test is deterministic and runs in polynomial time.
  • Encryption does not break. This is the most repeated myth. RSA is safe because no one knows a fast way to factor a large number into its two primes. RH says how primes are spread on average; it gives no method to factor anything. Elliptic-curve cryptography doesn't rest on RH either. Your bank, your messages, and Bitcoin work the same the day after.
  • An AI did much of the work. Alpöge credits Claude with the step that closed the argument, and the Lean proof was written by the two of them. Andrew Sutherland of MIT told Scientific American: "An AI that can prove the Riemann hypothesis is not one I'd want to meet." And: "If that happens, mathematicians having jobs will be the least of our problems."

Past claims that failed, and what is different now

YearWhoWhat happened
June 2004Louis de Branges, Purdue UniversityPosted a 124-page claimed proof on his university page. Conrey and Li showed the positivity conditions his approach needs fail, and the proof was not accepted
September 24, 2018Sir Michael Atiyah, at the Heidelberg Laureate ForumPresented a short claimed proof in a lecture. It did not convince experts

Both earlier claims had to wait for people to read them. This one arrives as a Lean proof: anyone can run the checker, and a third party is already running it. When Comparator finishes, the four questions in the table above have machine answers. People still read the statement and the definitions, and Alpöge's readable paper is still to come.

Who gets the $1 million prize

The Clay Mathematics Institute named the Riemann Hypothesis one of its 7 Millennium Prize Problems on May 24, 2000, with US$1,000,000 for each. Only the Poincaré Conjecture is listed as solved. Clay's rules set 3 conditions before it considers a solution:

  1. "the proposed solution must be published in a Qualifying Outlet";
  2. "at least two years must have passed since publication";
  3. "the proposed solution must have received general acceptance in the global mathematics community."

So the clock starts when the written proof is published, not when the Lean check passes. As of November 21, 2026, 01:50 UTC, Clay's page lists the problem as "Unsolved" and Clay has made no statement on the claim.

Make an explainer page with AI and share it as a link

Explaining this to a class, a team, or readers? Ask Claude or ChatGPT for one HTML page: the zeros on the critical line, a prime-counting chart against Li(x), and the 4 checks above. Then put it online with Convika:

  • From the chat. Add the Convika connector to Claude (setup) and say "publish this page on Convika." Claude saves the page and gives you a private preview link.
  • From the file. Sign up free, and on Put your HTML page online, upload the file or paste the code, then click Create my preview.

On the page's screen, click Publish page, choose Public, and click Next. The first time, click Publish free: the page is live at its own link, charts and buttons working. Send the link anywhere, and Convika shows how many people opened it. Pro ($12 a month) takes the Convika bar off.

Frequently asked questions

Has the Riemann Hypothesis been proven?

A proof is claimed and its full machine check is still building. On November 20, 2026, Levent Alpöge said he and Claude proved it, and published a Lean proof. An independent check passed its first gates, and MIT's Andrew Sutherland compiled the proof in Lean on his own. The full Comparator run had no result yet, and the Clay Mathematics Institute lists the problem as unsolved, as of November 21, 2026, 01:50 UTC (10:50 JST).

Did Claude solve the Riemann Hypothesis?

Alpöge says Claude did much of it. He writes, "working with Claude," and "this time Claude far outpaced me near the finish line," and credits Claude with the key step: a chain of theta lifts of length √X standing in for multiplicity. Claude and Alpöge also wrote the Lean formalization together. The note names Claude, not a specific model.

What happens if the Riemann Hypothesis is proven?

The error in counting primes is as small as it can be (von Koch, 1901), gaps between primes get a bound, and every theorem proved "assuming RH" or "assuming GRH" holds unconditionally. The Clay prize of US$1,000,000 becomes possible once the written proof is published and generally accepted for 2 years.

Does proving the Riemann Hypothesis break encryption?

No. RSA rests on how hard it is to factor large numbers, and RH gives no way to factor them. A proof says primes are spread as evenly as possible; it says nothing about finding the two primes inside a given number. No encryption breaks.

Who is Levent Alpöge?

A mathematician at Anthropic who works in number theory. In August 2026 he and Ralph Furman posted a paper showing that more than two thirds of the zeta zeros are simple and on the critical line, a proof the paper says Claude discovered autonomously, verified in Lean 4. On November 20, 2026, he announced the claimed proof of the Riemann Hypothesis.

What is the Generalized Riemann Hypothesis?

The same statement for Dirichlet L-functions, the cousins of the zeta function used to count primes in arithmetic progressions: no zero with real part above 1/2. Alpöge's Lean files state it for every modulus and every character.

How long does it take to check a Lean proof?

The full Comparator run compiles all 175,187 files, takes hours and about 100 GB of disk. As of November 21, 2026, 01:50 UTC (10:50 JST), it was still building, with no result yet.

Sources

Read on November 20, 2026, between 16:03 and 16:10 UTC, unless another read time is given. Items marked "read again" were read on November 21, 2026, between 01:46 and 01:55 UTC (10:46 to 10:55 JST).