The Riemann Hypothesis Proof Made With Claude Passes Lean Checks. The Clay Institute Still Lists It Unsolved
A proof of the Riemann Hypothesis, made by an Anthropic mathematician working with Claude, now has the strongest kind of check a proof can get from software. On Nov. 20, Levent Alpöge released it as a Lean library, and the Lean and Mathlib communities say they have certified it. The Clay Mathematics Institute still lists the problem as unsolved, and no human-readable version of the argument exists yet.
This is a fast-moving story, and the sources differ in how definitive they sound. Here is what is established, what is still running, and what is missing.
What was announced
Alpöge, who works at Anthropic, posted a four-page note and the formal proof on Nov. 20. His wording is careful: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” He adds that the method extends to the Generalized Riemann Hypothesis, expects it to reach the Extended version “with more time,” and that “Claude and I have also formalized the arguments without assumptions.”
He also writes that the argument is “incredibly long at the moment” and that he intends “to do my best to understand and impart understanding of this argument.” The proof is released as Lean code, which the community’s statement puts at 17 million lines.
What has been checked
- Lean and Mathlib communities. A joint statement, endorsed by the Lean FRO and the Mathlib Initiative, says: “We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.” It rests on three checks: the theorem Alpöge proved was compared with Mathlib’s own statement of the Riemann Hypothesis using two independent tools; dozens of people reproduced a correct build of the library; and the exported proof was accepted by several proof checkers, including Lean’s official kernel, run on an inert export of the proof.
- Andrew Sutherland (MIT). His verification page logs the checks. Sutherland notes the checks were run with Claude Code.
- Press. Scientific American reports that the result was verified by independent researchers.
| Check on MIT’s page | Status at last update |
|---|---|
| All 175,187 Lean modules compile | Pass |
| Five headline theorems use only three standard axioms | Pass |
| Independent sandboxed rebuild reproduces the published export | Pass, byte for byte |
| nanoda and con-leche (independent kernels) accept the full export | Accept |
| Lean’s default and paranoid kernels on the export | Still running |
| lake comparator, end to end | Still running |
What isn’t settled
- No human-readable proof yet. Terence Tao, in a Nov. 20 post, calls it an “(enormous) Lean-only proof” and writes: “I imagine the digestion process will take some time, though in this case at least we have a human author to communicate with.”
- Machine-checked is not the same as understood. The checks show the code proves the formal statement Mathlib calls
RiemannHypothesis. They don’t tell number theorists why the argument works. - Clay still says unsolved. Its Riemann Hypothesis page lists the problem as unsolved. Under Clay’s rules, a solution must be published in a qualifying outlet, at least two years must pass, and it must have received general acceptance.
- Checks were still running. At MIT’s last update, two kernel checks and the end-to-end comparator had not finished.
Tao’s post also lists two other AI-assisted results announced the same day, on blowup for the Euler and Navier-Stokes equations. We have not covered those here.
PCGeek’s take
For a tech audience, the interesting part is the checking. A proof this long can’t be refereed by reading it, so the community leaned on software: compile it, compare the statement to a vetted one, and have independent kernels replay it. That is a real test, and it is also a narrower claim than “mathematicians have accepted this.” The next milestones are the remaining checks and a write-up people can read.
Disclosure: Claude is made by Anthropic, which employs Alpöge, and Sutherland used Claude Code for some checks. We also use Claude in our own publishing workflow.
If a proof is too long for any person to read, does a machine check count as understanding? Tell us below. Follow @pcgeekinfo.
Art: “Complex zeta.jpg” by Jan Homann, public domain, via Wikimedia Commons. A color plot of the zeta function, cropped and recolored by PCGeek. It is not a picture of the proof. Reporting as of the night of Nov. 21, 2026 (PT).