
Anthropic Claims AI Proof of Generalized Riemann Hypothesis, Stanford Releases Human Draft
Anthropic mathematician Levent Alpöge and AI model Claude claimed a proof of the Generalized Riemann Hypothesis in Lean. Hours later, Stanford released a human draft by Alessandro Masullo. Both proofs face community review.
Anthropic mathematician Levent Alpöge announced on Friday that he and the AI model Claude have discovered a proof of the Generalized Riemann Hypothesis. The proof is published in the Lean programming language for computer verification. Hours later, Stanford University researchers released an independent draft proof by mathematician Alessandro Masullo that uses no AI.
AI-assisted proof and Lean verification
Alpöge said he had pursued the problem as a secret lifelong goal. Over recent months he worked closely with Claude, setting direction and combining approaches on a simplified model. The AI then took over and produced the key insight, finding that for a number sequence of length X the required chain of transformations equals the square root of X. The full argument would span over 250,000 pages if printed, so Alpöge released it only in Lean, split into 14 data packages. Lean checks each logical deduction against mathematical axioms, though human error could remain if the initial problem translation was flawed.
Stanford human draft and early review
Stanford professor and American Mathematical Society president Ravi Vakil posted a hurried announcement on his blog sharing Masullo's draft. Renowned number theorist Brian Conrad had reviewed the work quietly for months. The planned year-long refinement was cut short by Anthropic's announcement. Within eight hours Vakil updated the post noting that several mathematicians, aided by AI models, had identified problems in the draft. Vakil emphasized that all issues must be resolved before acceptance.
Significance and next steps
The Riemann hypothesis, open since 1859, underlies the distribution of prime numbers. A confirmed proof would validate hundreds of dependent theories and extend to the Generalized Riemann Hypothesis covering broader equations. Vakil, meanwhile, urged mathematicians to review Masullo's draft quickly, translate it into verification systems like Lean, and close any small gaps. Alpöge admits he has not fully mastered the surrounding math and plans a human-readable version. The episode follows a fast-moving 2026 in which Claude has solved several open math problems, raising questions about AI's role in research.
Sources: The Decoder · TechCrunch · Lobsters (Primary Pdf)
SiTech — AI-powered web development
We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.