November 21, 2026
Banner: Riemann hypothesis proof claimed with Claude AI, zeros glowing on the critical line

Image: TeQBlogs

An Anthropic mathematician has published a claimed Riemann hypothesis proof that he says was found while working with the Claude AI model, and written entirely in the Lean proof language. The code has reportedly compiled on an independent expert’s machine, but no outside mathematician has yet confirmed that the proof is correct, so treat this as a major claim under review, not a settled result.

Here is what was announced on 20 November 2026 (US time), what has and has not been checked, and why the answer matters well beyond mathematics.

Quick summary

  • What: a four-page note claiming a proof of the Riemann hypothesis, and of the wider Generalized Riemann Hypothesis (GRH), made “working with Claude”.
  • Who: a mathematician employed by Anthropic, posting on his personal website. It is not an official Anthropic announcement.
  • Format: the proof was released only as Lean code. The note says a human-readable version ran to more than 250,000 pages, so it was not published.
  • Status: unverified. Compiling the code is a strong first signal, but experts still need to confirm the formal statement says exactly what the hypothesis says.
  • Prize: the Clay Mathematics Institute’s US$1 million award needs journal publication, a two-year wait and general acceptance first.

What is the Riemann hypothesis?

The Riemann hypothesis is a 1859 conjecture that all the “interesting” zeros of a function called the Riemann zeta function lie on one vertical line. Because those zeros control how prime numbers are spread out, a proof would show that primes follow a hidden order rather than pure randomness.

Also read: GTA 6 Is Out in India: Price, Editions, Download Size and PC Status

The Clay Mathematics Institute lists it as one of its seven Millennium Prize Problems and notes that the first 10 trillion of these zeros have already been checked by computer. Checking examples is not a proof, though: a single zero off the line, however far out, would break the hypothesis. A large body of published research is written in the form “if the Riemann hypothesis is true, then…”, so a genuine proof would turn those conditional results into proven ones at once.

What the Riemann hypothesis proof claim actually says

The note opens with the author saying he believes he has, working with Claude, discovered a proof of the Riemann hypothesis, and that the method extends to the Generalized Riemann Hypothesis. He also writes that he expects the Extended Riemann Hypothesis to follow with more time.

Three details in the note stand out:

  1. Claude did much of the final work. The author describes guiding the effort to a simpler “toy” setting, after which, in his words, Claude outpaced him near the finish line. The key idea he highlights adapts an older technique (Stepanov’s method) using tools from the theory of automorphic forms.
  2. Only Lean was published. A footnote says the machine-generated human-readable version passed 250,000 pages, and he chose not to post what he called “slop”.
  3. The timing is addressed openly. The note acknowledges that the announcement came in the same week as Anthropic’s stock market debut, says he is aware of the cynical reading, and rejects it.

It is worth stressing that Anthropic itself had not published any statement on the claim at the time of writing. The note lives on the author’s own site.

Has the proof been verified?

Not yet. One step has gone well, but the steps that decide whether mathematics accepts it are still ahead.

Lean is a proof assistant: it accepts a theorem only if every step follows strictly from earlier definitions and results. According to Scientific American, an MIT mathematician compiled the Lean code independently on 20 November and called it a historic moment. That is significant, because a compiling Lean proof is far harder to fake than a long paper.

But Lean only checks that the proof proves the statement it was given. Experts still have to read the formal statement and every definition behind it to make sure it really says “every non-trivial zero has real part one half” and not something subtly weaker. A small slip there would mean a correct proof of a different theorem. Early comments from mathematicians online say the formal statement does appear to be GRH, but that is not the same as a full review.

Checklist of Riemann hypothesis proof verification steps and their status on 21 Nov 2026
Image: TeQBlogs
CheckStatus (21 Nov 2026)Why it matters
Lean code publishedDoneAnyone can try to compile and inspect it
Independent compileReported by one MIT mathematicianConfirms the logic runs without errors
Formal statement reviewedEarly, informal comments onlyConfirms Lean proved the real hypothesis
Human-readable write-upNot published (250,000+ pages)Needed for mathematicians to understand the ideas
Peer-reviewed publicationNot yetFirst requirement for the Clay prize
General acceptanceNot yetTakes months to years

How this builds on Claude’s earlier maths result

This is not the first time Claude has been linked to the Riemann zeta function. In August 2026, Anthropic said an unreleased research version of Claude had raised the proven lower bound for the share of zeta zeros on the critical line from 41.6% to 67.2%. That result was reviewed by two Anthropic mathematicians and examined by two outside experts, and Anthropic also released a Lean formalisation.

At the time, Anthropic wrote that it did not expect those techniques to lead to a proof of the full hypothesis. The new claim uses a different route and, if it holds, moves the figure from “at least 67.2%” to “all of them”. It also lands in a year when AI labs have been racing on famous maths problems; we covered a related example in our report on DeepMind’s AlphaProof code checking Signal’s encryption.

Timeline from 1859 to the claimed Riemann hypothesis proof in November 2026 and next steps
Image: TeQBlogs

Why it matters, even if you are not a mathematician

If confirmed, this would be the biggest mathematical result an AI system has helped produce so far. That would change how research is done: proofs could arrive as machine-checked code before any human fully understands them, which the author himself describes as an unusual and uneasy position.

Video: Pexels

Some common worries are misplaced, though:

  • Your bank and UPI encryption are not suddenly broken. RSA and similar systems depend on how hard it is to factor large numbers. The Riemann hypothesis is about how primes are distributed on average and does not give a way to factor a specific number.
  • Some algorithms gain firmer footing. Certain primality tests and number-theory results that were only proven “assuming GRH” would become unconditional.
  • AI credibility is on the line. A confirmed proof would be a landmark for AI reasoning; a flaw in the formal statement would be an equally important lesson about checking AI output.

What it means for India

For Indian students and researchers, the practical message is that formal proof tools such as Lean are moving from a niche skill to a core one. Lean and its maths library are free and open source, and learning to read formal statements is exactly the skill needed to judge claims like this one.

For businesses and developers using AI tools, the episode is a reminder that even impressive AI output needs independent verification. The same applies at a smaller scale to code, reports and data produced with assistants such as Claude, ChatGPT or Google’s newly launched Gemini 4 Neon.

What happens next

  1. Hours to days: more independent compiles of the Lean code and close reading of the formal statement by Lean and number-theory experts.
  2. Weeks to months: attempts to extract, simplify and write up the key ideas so mathematicians can understand them. The author says he expects shorter proofs to appear.
  3. Years: publication in a peer-reviewed journal, then at least two years and general acceptance before the Clay Mathematics Institute would consider the prize.

The claim also arrives during a busy week for the company, which completed its record listing days earlier. You can read the financial side in our Anthropic IPO explainer (not investment advice).

Frequently asked questions

Has the Riemann hypothesis been proved?

It has been claimed, not confirmed. A proof written in Lean was published on 20 November 2026 and has reportedly compiled, but experts have not yet finished checking that the formal statement matches the real hypothesis.

Did Claude solve the Riemann hypothesis by itself?

No. According to the note, a human mathematician steered the work and Claude contributed heavily, including the key final steps. The claim is made by the mathematician, not by Anthropic as a company.

Will the $1 million Clay prize be paid?

Not soon. Clay’s rules require publication in a qualifying journal, at least two years after publication, and general acceptance by mathematicians before a solution is even considered.

Does a Riemann hypothesis proof break encryption?

No. The hypothesis describes the average distribution of primes and does not provide a method to factor the large numbers that RSA encryption relies on.

Bottom line

This is the most ambitious AI-assisted maths claim yet, and the fact that it comes as compiling Lean code makes it far more credible than the usual “I proved Riemann” announcement. It is still a claim. Wait for experts to sign off on the formal statement before treating the Riemann hypothesis as solved; we will update this story as verification progresses.

Sources

(Visited 1 times, 1 visits today)

About The Author

Leave a Reply

Your email address will not be published. Required fields are marked *