The Riemann Hypothesis — the most famous unsolved problem in mathematics, open since 1859 — may have a proof. On 20 November 2026, Levent Alpöge, a mathematician at Anthropic and a Harvard junior fellow, announced that he believes he has, working with Anthropic's AI model Claude, discovered a proof of the Riemann Hypothesis — and of its broader form, the Generalised Riemann Hypothesis.
This is a CLAIM, announced — not a confirmed proof. No human mathematician has yet read the full argument, and the mathematical community has not accepted it. What follows is what has been claimed, what machine checks have confirmed so far, and what is still to be verified.
Key facts
| Item | Details |
|---|---|
| Announcement | 20 November 2026, by Levent Alpöge (Anthropic mathematician, Harvard junior fellow), working with Claude |
| What is claimed | A proof of the Riemann Hypothesis and the Generalised Riemann Hypothesis |
| What was released | A 4-page note ("Remark") + a 14-part Lean archive (~17.4 million lines, 175,187 files). No conventional paper was published. |
| The problem | Posed by Bernhard Riemann in 1859; the Riemann zeta function's non-obvious zeros all have real part exactly 1/2; a Clay Millennium Prize problem ($1 million) |
| Machine checks | All 175,187 modules compile without errors (MIT's Andrew Sutherland); independent checkers nanoda, con-leche and lean4lean accept the proof; only Lean's three standard axioms used |
| Human verification | NOT done — no human has read the full argument (a written version would run 250,000+ pages); final comparator run still in progress at last check |
| Status | Claim announced; verification pending; Clay Mathematics Institute still lists the hypothesis as unsolved |
What was actually released
Alpöge did not publish a conventional paper. He released a short 4-page note sketching the argument, together with the full formal proof written in Lean — a computer proof assistant that checks whether every step follows from the definitions and earlier results. The Lean archive runs to about 17.4 million lines across 175,187 files. Alpöge has said the written argument is "incredibly long at the moment" — more than 250,000 pages — and that he intends to understand the argument better, simplify it and "write it properly".
What the machine checks show so far
The most detailed public record comes from Andrew Sutherland, a mathematician at the Massachusetts Institute of Technology, who independently compiled all 175,187 modules of the proof with zero errors. Two independently written proof checkers, nanoda and con-leche, accept the full exported proof, and the lean4lean checker accepted all 175,187 modules module-by-module. The five headline theorems depend only on Lean's three standard axioms. A statement-comparison check confirms the theorems match the published challenge. Sutherland told Scientific American: "the chance that anything is fishy here is approximately zero."
"If this fully checks out, it is clearly a historic milestone in mathematics," said number theorist James Maynard, a Fields Medal winner at the University of Oxford.
What is NOT done yet
Machine checks confirm that the code is logically valid — but a machine cannot confirm on its own that the code proves the right thing. No human has read the full argument, and mathematicians "don't even know why the theorem is true" yet, in the words of Kevin Buzzard of Imperial College London. The final comparator run — checking that the code does not redefine any symbols and is not exploiting an unknown bug in Lean's own checker — was still in progress at last check on 21 November 2026. Full human understanding of the argument is expected to take weeks.
The criticism
Not everyone welcomed the announcement as it was made. Andrew Granville, a number theorist at the University of Montreal, told Nature that Alpöge should have taken the time to prepare a clearer write-up before making the announcement, which came just days after Anthropic's initial public offering. "What he released is garbled," Granville said of the 4-page document, and he cited the downturn in the company's stock value as the reason for the timing. Alpöge addressed the timing question in his document: "There is an obvious cynical view, to do with the timing, but it is false." Anthropic did not respond to Nature's request for comment.
A separate claim — do not mix them
A second, separate claim appeared the same week: Stanford researcher Alessandro Maria Masullo proposed a human-developed proof of the Generalised Riemann Hypothesis. That draft has already drawn questions from mathematicians examining it. The two claims are independent — neither verifies the other.
What happens next
The mathematical community will now spend weeks — at least — checking the work: finishing the comparator run, confirming that the definitions in the code faithfully capture the standard mathematical objects, and building a human understanding of why the argument works. Until the community treats the argument as settled, the Clay Mathematics Institute still lists the Riemann Hypothesis as unsolved for prize purposes. The claim is announced and the machine checks are passing — but the proof is not confirmed yet.
FAQs
Has the Riemann Hypothesis been proved?
No — not yet. A proof has been CLAIMED by Levent Alpöge, working with Claude, and announced on 20 November 2026. Machine checks of the Lean code are passing, but no human has read the full argument and the mathematical community has not accepted the proof.
What did Alpöge actually release?
A 4-page note sketching the argument, plus the full formal proof as a Lean archive of about 17.4 million lines across 175,187 files. No conventional paper was published.
What have the machine checks confirmed?
All 175,187 modules compile without errors, independent checkers (nanoda, con-leche, lean4lean) accept the proof, and the headline theorems depend only on Lean's three standard axioms. MIT's Andrew Sutherland put the chance that "anything is fishy" at "approximately zero".
What is still to be checked?
The final comparator run (confirming the code does not redefine symbols), confirmation that the code proves the real Riemann Hypothesis and not a subtly different statement, and full human understanding of the argument — expected to take weeks. Until then, the proof is a claim, not a confirmed result.
A note on verification
The announcement, the machine-check results and the expert reactions above are as reported by Nature, Scientific American, Al Jazeera, phys.org and Moneycontrol, and from MIT mathematician Andrew Sutherland's public verification page. The Granville criticism is attributed to Nature's reporting. The status given here (claim announced, verification pending) was last checked on 21 November 2026 and will change as verification proceeds — this post will be updated when the community's verdict is known.
Source: Nature; Scientific American; Al Jazeera; phys.org; Moneycontrol; A. Sutherland's verification page. Last checked: 21 November 2026.
Comments (0)
Share your question or experience about this syllabus. Your comment will appear after review.
No comments yet — be the first to share your thoughts!