Levent Alpöge, a mathematician who works at Anthropic, said on November 20, 2026 that he believes he has proved the Riemann hypothesis with the help of Claude, the company's AI model. In a four-page note on his own website he wrote that the method also settles the generalized Riemann hypothesis and, with more time, should reach the extended version. For now this remains a claim: it has not been verified independently and has not been through peer review.
What he released is a formal version of the argument written in Lean, a proof language in which a computer checks every logical step. Alpöge says Claude was integral to both the mathematics and the formalization. He is publishing only the Lean version at this stage, because a complete written proof would be extremely long and he has not yet mastered all of the mathematics involved. He says he intends to write it up properly. Scientific American reported that MIT mathematician Andrew Sutherland independently compiled the result in Lean on the day of the announcement.
The hypothesis was posed by Bernhard Riemann in 1859 and concerns how the prime numbers are distributed. It is one of the seven Millennium Prize Problems, each carrying a $1 million prize from the Clay Mathematics Institute. Under the institute's rules a solution must be published and withstand two years of scrutiny, so no prize decision is close. Only one of the seven, the Poincaré conjecture, has been officially resolved. In September OpenAI said an AI model had cracked another, the Navier-Stokes problem, and mathematicians describe that work as apparently correct but not yet fully vetted.
Reaction from specialists has been intense but careful. Brian Conrey, a leading expert on the hypothesis, told AFP it could be the biggest thing ever to happen in mathematics, while adding that he had seen only Alpöge's remarks and had asked for a manuscript. TechCrunch described number theorists working through the argument in a Zulip chat, where Chris Birkbeck of the University of East Anglia observed that a key bound rests on an automorphy-lifting argument. Most participants appeared confident the Lean result would hold, TechCrunch reported, although humans will need time to understand why.
Several questions are open. According to The Verge, Anthropic has not commented, and Alpöge says he texted chief executive Dario Amodei before posting. The release came a day after Anthropic's shares began trading on Nasdaq, a timing Alpöge acknowledged could be read cynically. In August the company had said a research version of Claude improved a long-standing result related to the hypothesis, while stating it did not expect the technique to prove it. Scientific American notes that issues can arise in Lean, so reviewers must confirm that the code states the hypothesis correctly as well as compiling.
A human-only attempt has surfaced too. TechCrunch reports that Stanford professor Ravi Vakil posted a draft proof of the generalized hypothesis by postdoctoral researcher Alessandro Masullo, who spent about twelve years on it, and described the work as entirely human, with no AI. Vakil said the draft was released early because of the news from Anthropic. Until experts finish checking both claims, the Riemann hypothesis should be treated as unresolved.