Anthropic mathematician claims Claude helped prove the Riemann hypothesis — but no one has verified it
Levent Alpöge, a mathematician on Anthropic's technical staff, announced Friday that he believes he has, working with Claude, discovered a proof of the Riemann hypothesis — the 167-year-old problem on prime numbers. Outside mathematicians have rebuilt the code and the checks are passing, but no human has read the proof and the Clay Mathematics Institute still lists the problem as unsolved.

The News
Levent Alpöge, a mathematician on Anthropic's technical staff, announced on Friday, November 20, on his personal website that he believes he has, working with Claude, discovered a proof of the Riemann hypothesis — the 167-year-old problem on the distribution of prime numbers, first posed in 1859 — and of the generalised Riemann hypothesis. Rather than a conventional paper, he released the argument as Lean code — about 17 million lines, in a 14-part archive — meant to be machine-checked step by step. He says the written argument would run to more than 250,000 pages.
By November 21, outside mathematicians had rebuilt the code. MIT mathematician Andrew Sutherland reported 175,187 modules processed, with the lean4lean module pass (175,187 modules) and a separate checker called nanoda having passed — the full Lean kernel replay and the final comparator run are still running — and leading number theorists were reported to consider the claim very probably right. But no human has read and understood the proof, a final comparator run was still pending, and the check that the code's definitions match the intended hypothesis was still under way.
The Clay Mathematics Institute still lists the problem as unsolved. This is a serious, machine-checked claim, not an accepted theorem — and the claim is Alpöge's personal statement, not an Anthropic announcement. Alpöge says he contacted Anthropic chief executive Dario Amodei for permission to announce as the company prepared its initial public offering. For context: in August 2026 Anthropic reported that Claude had raised the guaranteed minimum share of zeta zeros known to lie on the critical line from 41.6% to 67.2% — explicitly not a proof.
The machine checks are passing — but no human has read the proof, and the problem is still listed as unsolved.
Why It Matters
The claim matters because the Riemann hypothesis is one of the most famous unsolved problems in mathematics — a statement about how prime numbers are distributed, first posed in 1859, with a $1 million prize from the Clay Mathematics Institute for a proof. If it were proved, it would be one of the great mathematical events of the century, and the fact that the proof was produced working with an AI model would be a second historic event on top of the first.
It matters because of how the proof is being checked. A 17-million-line argument that would run to more than 250,000 pages on paper cannot be read by any human — so it is being checked by machine, with the lean4lean module pass and the nanoda checker having passed so far. The 175,187 modules processed so far, with those checks passed, is a real signal — but it is a signal about the code, not yet about whether the code proves what it claims to prove.
But the gap between the code checks out and the theorem is proved is exactly why the verdict is Mixed, not Bullish. No human has read or understood the proof, the final comparator run is still pending, and the check that the code's definitions match the intended hypothesis is still under way. Until those steps complete, this is a claim with promising machine checks — not a theorem, and not an Anthropic announcement.
Analyst's Take
Our call is Mixed. The machine checks passing on 175,187 modules is genuinely impressive, and leading number theorists are reported to consider the claim very probably right. But very probably right is not a proof, and the Clay Mathematics Institute still lists the problem as unsolved.
What to watch next is the final comparator run and the definition check — the step that confirms the code's definitions actually match the Riemann hypothesis as mathematicians understand it. Those are the two steps that stand between a machine-checked claim and an accepted theorem, and either could still find a problem.
For investors, the read-through is Anthropic, ticker ANT — but only as a read-through. The claim is by an Anthropic employee using Claude, and Anthropic itself made no announcement. If the proof holds up, it would be a remarkable demonstration of what Claude can do; if it does not, it is a reminder that a machine check is only as good as the definitions it checks. Reader poll: Will the proof hold up — yes or no? Vote: Yes, verified, No, flaw found, Too soon to tell.
Are you bullish or bearish on this story?
By The Numbers
FAQ
What is the Riemann hypothesis?
The Riemann hypothesis is a statement, first posed in 1859, about how prime numbers — numbers divisible only by one and themselves — are distributed. It is one of the most famous unsolved problems in mathematics, with a $1 million prize for a proof.
What is Lean?
Lean is a computer program that checks mathematical proofs step by step. Instead of publishing a paper, Alpöge released his argument as Lean code — about 17 million lines — so that computers can verify each step mechanically.
What does machine-checked mean?
Machine-checked means a computer program has verified each step of an argument, rather than a human reading it. The checks by Lean's kernel and a separate checker called nanoda have passed so far — but no human has read or understood the proof.
What is the Clay Mathematics Institute?
The Clay Mathematics Institute is a private organisation that offers a $1 million prize for solving each of seven great unsolved maths problems, including the Riemann hypothesis. It still lists the problem as unsolved.
👁
Source: IBM