Levent Alpöge, a mathematician at Anthropic, released a proof of the Riemann Hypothesis on November 20, 2026, in a note he published himself, describing it as the result of work with Claude. The proof takes the form of a library written in Lean, a proof assistant. On November 21, the Lean and Mathlib communities issued a joint statement certifying that both the statement and the proof are correct.

The Proof Is About 17 Million Lines of Lean, With No Human-Readable Paper Yet

Alpöge published a Lean library distributed as 14 split compressed archive files, along with a short four-page note. The Lean community has also set up a verification repository on GitHub that mirrors the submitted source and points to Alpöge's note as the original announcement. According to the joint statement, the library runs to about 17 million lines. Records published by researchers who took part in the checking put the number of modules at 175,187.

In the note, Alpöge writes that he believes he discovered a proof working with Claude. He says the method extends further and also resolves the Generalized Riemann Hypothesis (GRH), and he expects that, with more time, it will reach the Extended Riemann Hypothesis as well.

At the same time, the argument is extremely long and there is not yet a paper humans can read. Alpöge explains that he considered it inappropriate to post a massive, unreadable document, so he chose to present only the Lean code for now. The note also includes a timeline of the work compiled by Claude. According to it, the effort began around October 6, 2026, many of the pieces came together on November 15, and the GRH proof was reached on November 16.

Alpöge Credits Claude With the Decisive Idea

Alpöge writes that while he set the direction of the research, Claude far outpaced him near the finish. As an example, he points to the challenge of creating a notion of “multiplicity” over the integers, which a classical line of argument requires. Claude, he says, proposed the decisive length for chaining together an operation known as a “theta lift.”

He also acknowledges that he does not yet fully understand the parts that rely on areas outside his expertise, such as automorphic forms on higher-rank groups. He expects the proof to be greatly simplified and says he intends to understand the argument and write it up properly as a paper.

The Lean Community Checked Both the Statement and the Proof

The joint statement, published on November 21, is endorsed by the Lean FRO and the Mathlib Initiative. First, the communities used two comparison tools to confirm that the theorem proved by the library matches the statement of the Riemann Hypothesis as defined in the Mathlib math library. They then checked the proof itself with Lean's official kernel (the core component that performs the final check of a proof) and several independently developed checkers, all of which accepted it.

Andrew Sutherland of the Massachusetts Institute of Technology (MIT) and Michael Rubinstein of the University of Waterloo each published records of their own independent verification. According to those records, all 175,187 modules compile without errors, and the proof depends only on Lean's three standard axioms. The statements they checked also include GRH for Dirichlet L-functions.

However, the MIT record shows that some checks were still running at the time of writing. GRH for other kinds of L-functions is not covered by the checks. What the community statement certified as correct is the statement and proof of the Riemann Hypothesis.

Prize Consideration Is Still a Long Way Off

The Riemann Hypothesis is closely tied to the distribution of prime numbers and is one of the Millennium Prize Problems selected by the Clay Mathematics Institute. Under the institute's rules, however, a proposed solution is considered only after at least two years have passed since its publication in a qualifying outlet and it has gained general acceptance in the mathematics community. Because there is not yet a human-readable paper, meeting those conditions is expected to take time.

Alpöge released the proof under his own name. As of this writing, Anthropic itself has made no announcement about it on its news or research pages. In August 2026, the company announced a separate result in which Claude raised the lower bound on the fraction of zeros of the Riemann zeta function lying on the critical line from 41.6% to 67.2%, naming Alpöge as one of the company mathematicians who examined that work. At the time, the company said it did not expect those techniques to lead to a proof of the Riemann Hypothesis.

Our Take

It is significant that a proof of about 17 million lines passed several independent checks within roughly a day of its release. It shows that even a proof far too long for humans to verify by reading can be judged correct or incorrect quickly by machine checking.

That said, a statement and proof being formally correct is a different matter from mathematicians understanding what is inside. Alpöge himself admits he does not fully understand parts of it, and the work of turning the argument into a form people can read is still ahead. Even so, a workflow in which AI supplies a decisive research idea and machines verify that it is correct could influence how research and development are done beyond mathematics.

Summary

An Anthropic mathematician released a Lean proof of the Riemann Hypothesis developed with Claude, and the Lean community certified the statement and proof as correct. For people who use AI in research and development, it offers an early example of pairing AI with formal verification. The writing of a human-readable paper and the results of the remaining checks at MIT are the next things to watch.