Levent Alpöge, a mathematician at Anthropic, publicly claimed on November 20, 2026, that he and Claude had discovered a proof of the Riemann Hypothesis. His announcement also claims a result for the Generalized Riemann Hypothesis for Dirichlet L-functions, extending the proposed breakthrough beyond the classical problem.
The public record includes a short announcement, a proof sketch and links to a multipart Lean formalization. Nature reported the claim while emphasizing the need for rigorous checking. Scientific American separately reported that mathematician Andrew Sutherland had independently compiled Alpöge’s result in Lean.
Those are significant developments, but they leave different questions open. Compilation is a concrete technical check; acceptance of a proof also requires confirming exactly which mathematical statements were encoded and what the argument depends on. The news is an extraordinary, attributable proof claim with some reported checking, not an established solution.
What Alpöge Says He and Claude Proved
Alpöge’s announcement covers two related statements. The classical Riemann Hypothesis concerns the zeros of the Riemann zeta function, a mathematical function closely connected to the distribution of prime numbers. It says that every nontrivial zero lies on the line whose real part is one-half in the complex plane.
That location matters because the zeros govern how far the number of primes below a given threshold can depart from its predicted value. A proof would sharply constrain that error. It would reveal structure in the distribution of primes, without turning them into a simple, evenly spaced sequence.
The claimed generalization concerns Dirichlet L-functions, which provide related information about primes in arithmetic progressions. These are sequences formed by repeatedly adding a fixed number. The scope matters: a claim about Dirichlet L-functions should not be casually expanded into a claim about every broader version of the Riemann Hypothesis.
State of Proof’s intake record makes that distinction explicit. It records claims for classical RH and Dirichlet GRH, while describing the Extended Riemann Hypothesis as a possible future extension.
The released material also needs a precise description. Nature describes a four-page document containing a proof sketch. State of Proof records links to 14 Lean archive parts, with a prose write-up described as future work. Readers therefore should not expect the short announcement alone to supply a conventional, fully developed explanation of the mathematics.
My view is that this distinction deserves more attention than the announcement’s length. A short summary can introduce a substantial formal proof, but it puts more weight on access to the formal artifacts and on reviewers’ ability to connect them to familiar mathematics.
Claude’s Reported Role Goes Beyond Writing Assistance
Alpöge attributes substantive mathematical contributions to Claude. According to State of Proof’s account of his announcement, he identified and guided an approach based on an analogy with the finite-field setting. Claude supplied key steps, including an idea about how many theta lifts to chain together.
Theta lifts are a specialized part of the proposed argument. The announcement’s reference to their chain length is useful because it identifies a particular mathematical contribution attributed to Claude, rather than simply saying that AI “helped.”
Alpöge also credits Claude with collaborating on the Lean formalization. He says the arguments were formalized “without assumptions,” a claim recorded by State of Proof.
The division of labor remains Alpöge’s account, without an independent audit of which participant supplied each step. It supports reporting a claimed human–AI collaboration. It does not support describing Claude as having autonomously selected, solved and verified the problem.
That distinction matters for interpreting the AI achievement. Expert direction, mathematical discovery and formalization are different contributions. A validated proof could be a landmark for AI-assisted research while still depending heavily on the mathematician who chose the approach and guided the work.
The announcement also should not automatically be treated as an official Anthropic research release. The Verge reported that Anthropic had not commented or involved itself in Alpöge’s announcement. Nature likewise reported no immediate response to its request for comment.
Lean Changes the Verification Work, Not the Need for It
Lean allows mathematical definitions, statements and proofs to be expressed in a form that software can check. This can make a proposed proof much more inspectable than an informal assertion that all the details work.
But the decisive question is what the formal system has checked.
A formal argument can only establish the statement actually encoded, using the definitions and dependencies supplied. Reviewers must therefore inspect both the proof and the correspondence between its formal statement and the standard mathematical problem.
The available evidence falls into three categories:
| Public evidence | What it supports | What remains to be checked |
|---|---|---|
| Alpöge’s announcement | The author claims RH, Dirichlet GRH and a Lean formalization | Whether the argument establishes those claims |
| Reported independent Lean compilation | A separate mathematician reportedly compiled the result | The exact statements, assumptions and scope of that check |
| State of Proof’s intake record | A documented claim and a proposed examination plan | Its independent examination has not started |
Scientific American reports that MIT mathematician Andrew Sutherland independently compiled Alpöge’s result in Lean. That is more substantial than merely reading the announcement. The same report quotes Sutherland saying:
“There is much work to be done to understand the mathematical details of the proof”
State of Proof, meanwhile, explicitly labels its own status as examination not started. These accounts describe separate checking efforts. State of Proof’s status should not be interpreted as evidence that nobody else has examined any part of the release.
Its proposed next steps show what a substantive audit would involve:
- Identify the final declarations corresponding to classical RH and Dirichlet GRH.
- Compare definitions and quantifiers with the standard mathematical statements, checking that the formal claims cover the intended cases.
- Pin and inspect dependencies, so the exact imported material and assumptions are known.
Frequently Asked Questions
3 questions
1Has Claude Solved the Riemann Hypothesis?
Alpöge claims that he and Claude discovered a proof. The claim has not become an accepted solution, although Scientific American reports an independent Lean compilation.