Anthropic mathematician Levent Alpöge posted a statement on his personal website on Friday, Nov 20, saying the company's AI tool Claude helped him prove the Riemann Hypothesis, which has stood open since it was posed in 1859. The work took about six weeks. According to The Guardian, two teams have verified his claim. The same day, Lean formalization files for the conjecture appeared on the Lean proof assistant community blog and on an MIT page, setting off wide discussion on Hacker News.
The core claim: a six-week collaboration and a key insight
In his statement, Alpöge said Claude helped him develop a solution to the problem over roughly six weeks. According to X user rohanpaul_ai, Alpöge built on an approach he had conceived as a student: steering two ideas that worked on simplified "toy" versions of the problem toward the real thing. Claude then took over the advanced areas he was less familiar with and supplied the key insight.
Alpöge said Claude's ideas left him "stunned" and would make fellow mathematicians "fall off their chairs."
The Riemann Hypothesis is one of the most important unsolved problems in mathematics and has stood for 167 years since it was posed in the 19th century. If the proof holds, it would be a landmark in the history of mathematics.
Verification: two teams and Lean formalization
The Guardian reported that two teams have verified Alpöge's claim, but the report did not disclose the teams' identities or the details of their verification. Meanwhile, two related threads appeared on Hacker News: one pointing to a post on the Lean proof assistant community blog titled "Lean verification of the Riemann Hypothesis," which drew 640 points and 506 comments; the other pointing to Lean proof verification files on the MIT mathematics department website, which drew 61 points and 21 comments. Lean is an interactive theorem prover whose formal verification can check the correctness of a proof at the logical level.
Reaction and background
The claim has not yet undergone formal peer review by the mathematics community. X user rohanpaul_ai noted in relaying the news: "The proof has not yet been verified by peer review in the mathematics community; it is a blockbuster but unconfirmed claim." That caveat stands in contrast to the appearance of the Lean formalization files: formal verification can provide a logical check, but it is not the same as full peer review by mathematicians.
The same thread also noted that PayPal co-founder Peter Thiel predicted in 2024 that "the situation is much worse for people who study math than for people who study words," and that within three to five years AI would solve every US Mathematical Olympiad problem. Some observers see this episode as a landmark test of the limits of AI-assisted mathematical research.
What to watch
- The identities of the two verification teams, their methods and their conclusions have not been made public and await further disclosure.
- The completeness and reproducibility of the Lean formalization files, and the Lean community's review of them.
- Whether formal peer review by the mathematics community begins, and how mainstream mathematicians respond publicly.
- If the proof holds, its impact on research related to the Riemann Hypothesis and on the paradigm of AI-assisted mathematical research.
