Saturday, 21 November 2026

Trusted News Discovery – Powered by AI and Verified by Humans

Levent Alpöge claims an AI-found proof of the generalized Riemann hypothesis; it is still unverified

Levent Alpöge says Claude helped him prove the generalized Riemann hypothesis in Lean code, but the claim is unreviewed and still unverified.

Produced with AI assistance and reviewed by the Badiyaa Editorial Desk. How we work.

Grok Fact CheckCheck it yourself in one tapVerify with Grok ↗Ask @grok on X ↗
The Oberwolfach mathematics research institute in Germany photographed in 2020.
Image: Christoph Aistleitner, CC0 via Wikimedia Commons

Levent Alpöge, a number theorist who now works at the AI company Anthropic, says the Claude model helped him find a proof of the generalized Riemann hypothesis, and he has posted it not as a paper but as a vast body of computer-checkable Lean code. The claim, made on November 20, has not been accepted by the mathematical community, has not been peer reviewed and has no human-readable version yet.

If it holds up, it would settle a question that has stood since 1859 and one of the seven problems carrying a $1 million prize from the Clay Mathematics Institute. That is a very large if, and this piece treats it strictly as a claim.

What Alpöge says he has

According to Wikipedia's biography of Alpöge, his note says Claude found a proof of the original Riemann hypothesis that extends to the generalized version for Dirichlet L-functions. The same entry says a full prose write-up would run past 250,000 pages and that the argument remains unverified. Wikipedia's Riemann hypothesis article adds that work on the Lean files ran from October 6 to November 16.

Lean is a proof assistant: a program that accepts a statement only when every logical step can be checked mechanically. That is why a Lean release draws more attention than a typical bold preprint. It also shifts the questions people ask. Does the code compile? Does it rely on any hidden assumption? Do its definitions really say what the famous conjecture says? Wikipedia's account reports that some mathematicians have compiled the code, while others have objected that nobody yet understands the argument in human terms.

Alpöge is no outsider. His arXiv author page lists 16 papers in number theory and combinatorics, and his personal website describes a Princeton PhD under Manjul Bhargava and a junior fellowship at Harvard's Society of Fellows before he moved to Anthropic.

What the hypothesis actually says

Bernhard Riemann linked the spacing of prime numbers to a function now called zeta. As the Clay Mathematics Institute's problem page puts it, the hypothesis says all of the function's interesting zeros sit on one vertical line, and the first ten trillion of them have been checked by computer.

The generalized version asks the same thing of a whole family of related functions built from Dirichlet characters, which track primes in arithmetic sequences such as 3, 7, 11, 15 and onward. Wikipedia's article on the generalized hypothesis says Adolf Piltz probably first stated it in 1884. A great deal of number theory is already written as "if GRH is true, then...", including a guarantee that the Miller primality test runs quickly.

Putting the numbers in perspective

Some quick arithmetic shows why outside checking will take time. The Lean phase lasted 41 days, from October 6 to November 16. Spreading a 250,000-page argument over 41 days works out to roughly 6,100 pages a day, far more than any human team could read in that time.

The question has waited 167 years since Riemann and 142 years since Piltz. It was not the only attempt on November 20, either. Ravi Vakil, president of the American Mathematical Society, posted a separate, unfinished manuscript by Alessandro Masullo the same day, and Vakil later noted that readers had already found gaps in that draft.

Limits

Nothing here is settled. A Lean file that compiles proves only the statement written in the file, so reviewers must confirm that the formal definitions match the real conjecture. The Clay Institute has made no statement on the claim, and no journal has reviewed it. Badiyaa explained the background in our guide to the 1859 prime-number puzzle, and covered AI and formal checking in Anthropic's claim that Claude solved two black hole physics problems and DeepMind's machine-checked audit of Signal's encryption code.

The next concrete step is separate builds of the full Lean archive, with each team publishing the exact list of axioms the main theorem depends on.

Sources

Image: Christoph Aistleitner, CC0 via Wikimedia Commons

This article was produced with AI assistance and reviewed by our Editorial Desk. How we work.

Grok Fact CheckCheck it yourself in one tapVerify with Grok ↗Ask @grok on X ↗