Anthropic mathematician Levent Alpoge has published a four-page note claiming a Claude Riemann proof, saying he found it working with the company’s Claude models, and has released the work as machine-checkable Lean code rather than as a conventional paper. The announcement, made on Friday, November 20, by Levent Alpoge on his personal website, lands one day after Anthropic began trading on the stock market, and it has not been independently verified.
Claude Riemann proof: A note, not a paper
MLQ.ai reported that Alpoge wrote in the note that he believes he has, working with Claude, discovered a proof of the Riemann Hypothesis, the best-known open problem in mathematics. There is no paper attached. The note, titled “Remark” and hosted at alpo.ge, points readers to a Lean project split into 14 compressed archive parts on the same site. Alpoge writes that it being obviously inappropriate to attempt to post some massive unreadable travesty of slop, he decided to only present Lean. A footnote in the note says the informal, AI-generated write-up ran past 250,000 pages.
MLQ.ai reported that Alpoge expects to reach the Extended Riemann Hypothesis with more time, and that he and Claude have also formalized the arguments without assumptions in Lean, a programming language that checks every logical step of a proof by machine. He hedges in the note itself, saying he is presenting the world with a complete argument before having mastered a massive neighbourhood around its mathematics, that he may have baked lots of strange suboptimal choices into the argument, and that he intends to write it properly.
OfficeChai, as reported by MLQ.ai, noted that the note gives no build instructions, size or list of axioms, and that nobody outside has yet shown the archive compiles or that its final theorem faithfully states the hypothesis. That distinction matters. A Lean proof is only as good as the statement it proves and the assumptions it allows, so the first checks experts will run are mechanical: does the code compile, and does it actually state the conventional conjecture.
What the hypothesis says and why it matters
MLQ.ai reported that Bernhard Riemann proposed the hypothesis in 1859, that it concerns the zeta function, whose zeros encode how prime numbers are spread along the number line, and that the hypothesis says all of the function’s nontrivial zeros sit on a single vertical line in the complex plane, the critical line. New Scientist reported that the hypothesis was first put forward in 1859 by German mathematician Bernhard Riemann, and that the question is one of the Clay Mathematics Institute Millennium Prize Problems, which will award a 1 million dollar prize for an accepted solution. Scientific American reported that the conjecture has since become the North Star for the entire field of number theory, and that progress toward solving the Riemann hypothesis was so rare that most top experts spent little of their time actually working on it.
The stakes run well beyond one problem. MLQ.ai reported that thousands of results in number theory are proved on the assumption that the hypothesis is true, which is why a proof would matter well beyond number theory itself. Scientific American reported that Hector Pasten, a mathematician at the Pontifical Catholic University of Chile, said that if the proof is correct and understood, this completes a dictionary between different branches of math where the connections were only partial, something like a Rosetta Stone. Carlo Pagano, a professor at Concordia University in Quebec and a visiting researcher at Google DeepMind, said it is very plausible that the new ideas in the proof will open up brand new avenues of thinking in number theory.
James McKee at Royal Holloway University of London told New Scientist that there are numerous results that are conditional on the generalised Riemann hypothesis, and now those are theorems. McKee added that most mathematicians believed, or the more cautious ones merely suspected, that the generalised hypothesis was true, so its proof will not be unsettling. The unsettling part, he said, is the rapid progress in recent months towards automated proofs of significant mathematical results.
Early signs of life, and open questions
MLQ.ai reported that the note goes further, saying the method easily extends much further and resolves the so-called Generalized Riemann Hypothesis, which Interesting Engineering reported extends the original conjecture to related mathematical functions called Dirichlet L-functions. MLQ.ai reported that in August Anthropic said an unreleased research version of Claude had raised the known lower bound on the share of zeta zeros lying on the critical line from 41.6 percent to 67.2 percent, a result Alpoge and colleague Ralph Furman reviewed, and that Anthropic said then that Claude’s attempt at the full hypothesis had failed and that it did not expect those techniques to prove it.
AI Chat Daily reported that the August 10, 2026 result used two Claude Code sessions, about 31 million output tokens and roughly 60 coordinating subagents, and that Anthropic said explicitly at the time that it did not prove the hypothesis. AI Chat Daily also reported that the August result was examined by Alpoge and Ralph Furman, and that the paper was reviewed by number theorists Brian Conrey and Dan Goldston, while no comparable outside review of the new proof has been announced, and Anthropic has not issued a statement on the claim.
Verification has begun at the edges. TechCrunch reported that several participants in the Zulip chat, including MIT mathematician Andrew Sutherland, say they have successfully confirmed that Alpoge’s solution can be compiled successfully, and that most mathematicians appear confident that the formalized Lean result will hold. Interesting Engineering reported that an independent verification project has reportedly confirmed that the downloaded files match Alpoge’s published archive, and that its initial checks also found that the main theorem statements compile against Mathlib, Lean’s standard mathematics library. Interesting Engineering also reported that researchers must still examine the complete proof, establish that the code addresses the conventional Riemann hypothesis, check its mathematical definitions and identify any unsupported assumptions.
Kevin Buzzard at Imperial College London reacted to the news, New Scientist reported, by saying let’s see if it compiles first. He added that it might be a while before the relevant experts can tell us the thrust of the argument, assuming it is all true.
How the argument was put together
MLQ.ai reported that the approach adapts Stepanov’s method, which proved the Riemann hypothesis for curves over finite fields, and that the note says Claude suggested building that multiplicity from a long chain of theta lifts, with the chain’s length growing like the square root of the size of the primes being counted. AI Chat Daily reported that the approach builds on Stepanov’s method from the function-field setting, and that reproducing that method’s multiplicity effect over the integers was the central obstacle. Interesting Engineering reported that the method uses weighted prime sums, known as von Mangoldt sums, along with mathematical constructions called theta lifts.
MLQ.ai reported that a timeline Claude compiled dates the first steps to October 6 and the final step, the Generalized Riemann Hypothesis, to November 16. MLQ.ai reported that Alpoge credits Claude with much of the work near the end, writing that this time Claude far outpaced me near the finish line, and that he personally has little mastery of automorphic forms on higher rank groups, the area where Claude did much of the technical work. He said he guided us to the right situation before Claude began to leave him in awe, and that some descriptions of those steps paraphrase an AI-written summary that might make zero sense.
MLQ.ai reported that the note thanks mathematicians including Manjul Bhargava, Peter Sarnak and Maksym Radziwill, and that being thanked does not mean they have checked the proof. Crypto Briefing reported that Alpoge earned his PhD from Princeton in 2020, working under advisor Manjul Bhargava, and that experts have stressed that the Lean formalization needs careful, independent scrutiny before anyone treats the claim as credible, because a machine can confirm that a proof is valid while humans still have to confirm that the proof is of the right thing. AI Chat Daily reported that Alpoge is also a junior fellow at Harvard’s Society of Fellows, and that the Lean project was released as 14 compressed archive parts, a package reported at roughly 175,187 source files and 239 MB.
Why the prize is not close
MLQ.ai reported that the Clay Mathematics Institute lists the hypothesis among seven Millennium Prize Problems with 1 million dollars allocated to each, that under Clay’s rules a solution must appear in a qualifying publication, at least two years must pass, and it must win general acceptance in the global mathematics community, and that the institute does not accept direct submissions. The Hindu reported that the Clay Mathematics Institute selected the seven Millennium Prize Problems in 2000 and offers 1 million dollars for a solution to each, and that under the institute’s rules a solution must be published in a peer-reviewed journal and withstand two years of scrutiny. Interesting Engineering reported that Alpoge’s announcement does not qualify for immediate prize consideration, and that until independent researchers validate the proof, the 167-year-old problem remains officially unsolved. The Hindu reported that the claim has not been independently verified and that the proof has not yet been reviewed by outside experts.
Timing is already part of the story. New Scientist reported that Anthropic listed on the stock market on November 19, and that Alpoge appears to suggest that he waited to announce the proof until after. MLQ.ai reported that the note appeared a day after Anthropic’s shares began trading on Nasdaq, and that Alpoge addresses this directly, writing that the precise timing is horrible, that the result came together late Sunday night, that he messaged Dario about his ability to announce, and that there was a constraint to respect. TechCrunch reported that Alpoge described how he messaged Anthropic CEO Dario Amodei last Sunday night to ask how to announce that he had found a Lean-compilable solution to the Riemann hypothesis.
There is also a human counter-current running in parallel. TechCrunch reported that Ravi Vakil, a Stanford professor and the president of the American Mathematics Association, posted a proof of the General Riemann hypothesis written by Stanford postdoc Alessandro Masullo, noting that it is all human, with no AI. TechCrunch reported that Masullo said the result is the work of 12 years, and that without being expressed in Lean, it is too early to say whether the proof checks out. MLQ.ai reported that OpenAI separately released a Quasi-Riemann Hypothesis result about six weeks ago that stopped short of the full hypothesis, and The Hindu reported that the claim comes two months after rival OpenAI said an AI model had solved another Millennium Prize Problem, on the equations that govern how fluids move.
For now, the burden sits with verification, not celebration. The evidence so far is a machine-checked formalization that several mathematicians say compiles, a package whose files appear to match what was published, and a set of caveats from the claimant himself. Whether the announcement becomes a milestone in computational mathematics or a footnote about overclaiming depends on what independent readers find inside the archive. This is a Claude Riemann proof in form, in Lean code and in timeline, and its meaning will be settled by people who can read mathematics, not by the systems that wrote it.

