Levent Alpöge says he and Claude proved the Riemann hypothesis in a Lean proof posted on 20 November. Experts are now checking the unverified claim.
Levent Alpöge, a mathematician at AI company Anthropic, said on Friday, 20 November, that he has used the company's Claude model to find a proof of the Riemann hypothesis, the most famous unsolved problem in mathematics. In a four-page note posted on his personal website, he wrote that he believes he discovered the proof working with Claude, and that the same method also settles the Generalized Riemann Hypothesis. Rather than a conventional paper, he released the argument as a formal proof written in Lean, a programming language that lets a computer check every logical step. AFP reports that a full written version would be extremely long, and Alpöge says a human-readable account will follow later.
The hypothesis, set out by Bernhard Riemann in 1859, predicts where the zeros of the zeta function lie, and it underpins much of what mathematicians know about how prime numbers are spread out. It is one of the seven Millennium Prize Problems, each carrying a $1 million award from the Clay Mathematics Institute. The claim has not been independently verified or peer-reviewed. Number theorist Brian Conrey told AFP it would be the biggest event in the history of mathematics if confirmed, though he had not yet seen a manuscript. Others urged caution, noting that experts must first confirm the Lean statement really matches the hypothesis. The announcement also came a day after Anthropic's record stock market debut, timing Alpöge admitted could be viewed cynically.
Checking has already begun. Lean specialists are trying to download the code, which is split into 14 archive parts, build it and confirm that the final theorem uses only standard assumptions, according to OfficeChai. Even if the code passes, mathematicians will want to understand why the argument works, and Alpöge has said he does not yet master all of it and expects shorter proofs to emerge. Anthropic did not immediately respond to AFP's request for comment. Earlier this year an unreleased Claude model raised a known lower bound on the share of zeta zeros on the critical line from about 41% to 67%, a result the company stressed was not a proof. Under Clay rules, any prize would require journal publication and two years of scrutiny.



