A Claude-assisted Lean result and a human draft put verification at the center of AI’s biggest math claim.
On Nov. 20, Levent Alpöge claimed a Claude-assisted proof of the Generalized Riemann Hypothesis. The Anthropic mathematician released Lean code for a problem that would reshape number theory. The original Riemann hypothesis, proposed in 1859, concerns the distribution of prime numbers.
Claude far outpaced me near the finish line.- Levent Alpöge, Anthropic mathematician
MIT mathematician Andrew Sutherland told Scientific American he independently compiled the Lean result. That is substantial evidence for the formal argument. Experts still need to check the theorem’s meaning and understand its mathematics. Compilation is a milestone. Mathematical review continues.
Alpöge’s timeline begins Oct. 6 and ends Nov. 16 with “GRH.” He says Claude outpaced him near the finish. He released only Lean because the argument was enormous and he had not mastered all its mathematics. A written-out version passed 250,000 pages, according to his note.
Hours later on Nov. 20, Stanford’s Ravi Vakil released Alessandro Maria Masullo’s separate human draft. Brian Conrad had examined it for
months. Conrad told Scientific American it was undergoing major revision. About eight hours later, Vakil reported problems identified with help from different AI models.
On Aug. 10, Anthropic had published narrower progress on the Riemann zeta function. A research version of Claude raised a lower bound from 41.6% to 67.2%. The lab said it did not expect those techniques to prove Riemann. Friday’s claim reaches far beyond that earlier result.
The release followed Anthropic’s Nov. 19 Nasdaq debut. Alpöge calls the timing “horrible” and says he consulted CEO Dario Amodei about announcing it. Both claims now face mathematical scrutiny, with a compiled artifact on one side and a revisable draft on the other. The proof claim arrived before the explanation.
