知Yizhi Daily
← Back to this edition

AI

Anthropic employee says Claude helped prove the Riemann hypothesis

Edition 2026-11-21 · 08:00 Beijing time

Anthropic employee and mathematician Levent Alpöge says that, with assistance from a non-public version of Claude, he completed a proof of the Riemann hypothesis in about six weeks and formalized it in Lean. The proof is more than 250000 lines long and has yet to be independently verified; prominent mathematician Terence Tao says digesting and checking it will still take time. The announcement came the day after Anthropic’s IPO and extends a trend of AI companies intensively claiming to have conquered difficult mathematical problems.

According to BBC World, Alpöge announced the claim on Friday, one day after Anthropic listed on the U.S. stock market. He said he had asked company CEO Dario Amodei on Sunday how to release it, and described the timing as less than ideal; if it had been completed earlier, it would not have coincided with the listing. He rejected the possible perception that the release was tied to IPO marketing.

The Riemann hypothesis is nearly 170 years old. It concerns the distribution pattern of prime numbers along the number line and is widely regarded as one of the most famous unsolved problems in mathematics. The complete proof submitted by Alpöge was written in Lean, a computer language used to verify mathematical proofs, and exceeds 250000 lines. If the conclusion holds, it would be a major breakthrough in the history of mathematics, but it remains pending verification for now.

Mathematician Terence Tao said verifying the proof will still take quite a long time, though this time there is at least a human author who can communicate and help with understanding. The report also noted that in September this year, OpenAI said employees had partially solved a difficult Navier-Stokes-related problem in 88 hours using non-public AI tools, involving more than 10000 AI agents. That proof likewise remains unconfirmed and drew criticism from academia about AI solving mathematical problems.

Original sources

BBC World ↗
Back to this editionOpen interactive reader