ResearchNews
Mathematician claims Riemann proof made with Claude
Anthropic mathematician Levent Alpöge says he proved the Riemann Hypothesis working with Claude. Independent kernels have accepted the formal Lean proof, but no human has yet understood it.
· 2 min read

What happened?
On November 20, Levent Alpöge, a mathematician at Anthropic, posted a four-page note on his website (Alpöge): "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." According to Science (Science), the AI-generated explanation of the proof runs over 250,000 pages and has not been released. Instead he published 17 million lines of Lean code that formally verify the generalized Riemann hypothesis too.
The Riemann Hypothesis, open since 1859, concerns the distribution of prime numbers and is one of the Clay Institute's seven Millennium Prize Problems. Science reports that Alpöge also released results on the Euler and Navier-Stokes equations.
How solid is it?
Formal checking moved fast. According to checks run by Andrew Sutherland (MIT) with Claude Code (MIT):
- all 175,187 Lean modules compile without errors;
- the five headline theorems depend only on the three standard axioms;
- the Lean kernel replayed every module with zero rejections;
- two independent kernels (nanoda and con-leche) and lean4lean accept the proof;
- the theorem proves Mathlib's own RiemannHypothesis definition, not a restated version.
Most mathematicians Science spoke to expect the proof to hold up. Ghaith Hiary of Ohio State University said its "blast radius extends to the whole of mathematics." There is no formal peer review or Clay Institute decision yet.
The catch
The code is larger than some operating systems, and mathematicians say it could take years before humans can digest it. Maksym Radziwill of NYU noted that the point of mathematics is understanding why things happen, not just collecting true statements. Sources told Science the compute likely cost millions of dollars; Alpöge has not disclosed a figure.
Why it matters
If it holds, this is the first time a centuries-old flagship problem has been solved with AI in machine-checkable form. The lesson for developers: formal verification (Lean, independent kernels) is what makes an AI-produced, humanly opaque result credible. The same pattern may come to code, with generated code checked by proofs rather than tests alone. For Anthropic, it lands a day after its record IPO (Anthropic IPO: ANT closes day one 17% above its $83 price).
Produced with the help of AI, reviewed by an editor. How we make our articles
Reviewed and edited by the Neural Daily desk. About


