Claude Riemann Hypothesis: AI Proof Verified by Experts
An Anthropic mathematician says he has solved the most famous unsolved problem in mathematics — with the help of AI. Levent Alpöge announced on Friday, November 20, 2026, that he believes he has, working with Claude, discovered a proof of the Riemann Hypothesis. A day later, the mathematicians who maintain the proof-checking software said the proof checks out.
If the claim holds up, it is one of the biggest moments in the history of mathematics — and the clearest sign yet that AI can do frontier research, not just routine work. Here is what was announced, how it was checked, and what still has to happen before anyone collects the $1 million prize.
What is the Riemann hypothesis?
The Riemann hypothesis was proposed by German mathematician Bernhard Riemann in 1859. It is about the distribution of prime numbers — the numbers divisible only by themselves and 1.
In simple terms, it says that all the non-trivial zeros of the Riemann zeta function lie on a single vertical line, called the critical line, where the real part equals one half. Mathematicians have checked trillions of these zeros by computer, and every one of them sits on the line — but nobody had ever proved that all of them must.
It is one of the seven Millennium Prize Problems set by the Clay Mathematics Institute, each carrying a $1 million prize. For 167 years, it has been widely regarded as the most important open question in pure mathematics.
What did Alpöge announce?
Alpöge, a mathematician at Anthropic and a junior fellow at the Harvard Society of Fellows, published a four-page note titled “Remark” on his personal website on November 20. In it, he wrote: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.”
The claim covers not only the classical Riemann hypothesis but also its broader version, the Generalized Riemann Hypothesis, for all Dirichlet L-functions.
What makes the announcement unusual is its form. Alpöge did not publish a conventional written proof. He released only computer code: a 14-part archive containing 175,187 files written in Lean, a programming language that lets computers check mathematical arguments step by step. The archive holds about 17.4 million lines of code, compressed to 239 megabytes and expanding to more than 61.5 gigabytes.
He said a normal written version of the argument would run to more than 250,000 pages, and that he has not yet mastered all of the mathematics involved himself. He plans to write it up properly when he is able to simplify it.
He also described how the work came together. He guided the high-level architecture, he said, while Claude handled complex algebraic manipulations and technical lemmas near the end. “Claude flew, and danced, and sparkled near that finish line,” he wrote.
The timing raised eyebrows: the announcement came the day after Anthropic’s stock market debut — our report on Anthropic’s Nasdaq debut covers the $81.3 billion listing, the largest IPO on record. Alpöge said he had consulted Anthropic chief executive Dario Amodei before publishing.
Has the proof been verified?
Yes — by machine-checking, and the checking was thorough. On Saturday, November 21, the Lean and Mathlib communities, which maintain the proof-checking software, issued a joint statement: “We have full confidence that Alpöge’s proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.”
Their certification rests on four checks. First, the statement of the hypothesis in the code is Mathlib’s own definition, carefully vetted by experts. Second, the whole library compiles without errors — dozens of people reproduced the build. Third, special comparison tools confirmed that the theorem proved in Alpöge’s code is exactly Mathlib’s statement of the Riemann hypothesis, so nothing was redefined along the way. And fourth, the exported proof was accepted by several independent checkers, including Lean’s official kernel and checkers written in other programming languages.
MIT mathematician Andrew Sutherland coordinated much of the verification. He reported that all 175,187 modules compiled with zero errors and that the proof uses only the three standard axioms of mathematics. Scientific American reports that independent mathematicians confirm the proof, and its headline calls the proof “almost certainly true”.

But there is an important limit to what has been verified. No human has yet read or digested the full argument — at 17 million lines of code, that will take a long time. Mathematician Terence Tao has said the community will need time to work through it.
That also means the $1 million prize is not settled. The Clay Mathematics Institute awards its prizes through its own process, which requires peer review and publication — a computer check, however rigorous, does not collect the prize on its own.
Why this matters
This is the first time a Millennium Prize Problem has been claimed through a fully compiled formal proof rather than a conventional paper. It also follows a striking run of AI results in mathematics: in September, OpenAI overtook Alpöge’s own team on the Navier-Stokes problem, another Millennium Prize Problem, and in July Alpöge announced a proof of a different famous conjecture.
The pattern is now hard to ignore. In three months, AI systems have been at the centre of claimed solutions to two of the seven Millennium Prize Problems. Whether or not every claim survives scrutiny, the way mathematics is done is changing — machines can now produce arguments far too long for any human to read, and the job of checking them is moving to other machines.
There is also a rival claim to watch. Stanford researcher Alessandro Masullo has proposed a separate, entirely human-developed proof of the Generalized Riemann Hypothesis. That draft is at an earlier stage of scrutiny, with mathematicians raising questions about it.
What happens next?
Mathematicians will spend months working through the ideas behind the code, and Alpöge plans to publish a human-readable write-up. The Clay Institute’s prize process will take its own course. And the proof itself will be picked apart, line by line, by people who want to understand not just whether it is right, but how it works.
For now, the careful way to put it is this: a machine-checked proof of the Riemann hypothesis exists, independent checkers say it is correct, and the mathematical world is only just beginning to understand it.
FAQ
Did Claude prove the Riemann hypothesis on its own?
No. Alpöge guided the high-level ideas and translated them into the right mathematical setting, while Claude carried out complex calculations and technical steps near the end. It was a collaboration, with the AI doing more of the work as the proof came together.
Is the proof officially accepted?
It has been machine-checked and certified by the Lean and Mathlib communities, but it has not yet been through human peer review. That process will take time.
What is the $1 million prize?
The Riemann hypothesis is one of seven Millennium Prize Problems named by the Clay Mathematics Institute in 2000. Each carries a $1 million prize for a correct solution that passes the institute’s review process.
What is Lean?
Lean is a programming language and proof assistant. Mathematicians write a statement and a proof in Lean, and a small program called the kernel checks every step against a short list of rules. If the code compiles without errors, the logic of each step is correct.


