An Anthropic mathematician says he and Claude proved the Riemann hypothesis. Now the experts check
Levent Alpöge posted a four-page note on his website claiming a proof, made with Claude, of the Riemann hypothesis and its generalised version. There is no paper: the proof is formalised in Lean across 14 archive files. MIT's Andrew Sutherland has already compiled it. Expert review of the mathematics has not happened yet.
Levent Alpöge, a mathematician on Anthropic’s technical staff, posted a four-page note on his personal website saying that, “working with Claude”, he has discovered a proof of the Riemann hypothesis. It went up on Friday. The problem was posed by Bernhard Riemann in 1859, 167 years ago, and is widely seen as the most important open question in mathematics.
The note claims more than that. Alpöge says the same method settles the generalised Riemann hypothesis, which covers a whole family of related functions, and that he expects to reach the “extended” version given more time.
What does the Riemann hypothesis actually say?
Prime numbers look as if they are scattered at random. The hypothesis says they are not: how they thin out is governed by the zeros of one function, the Riemann zeta function, and every zero that matters sits on a single vertical line, the critical line. If that holds, the primes are spread about as evenly as they possibly could be.
Thousands of theorems in number theory have been proved on the assumption that it is true. A proof would turn all of them into settled results at once.
Why is the proof only available as code?
There is no paper. Alpöge writes that the argument is enormously long and that he did not want to post a huge, unreadable document. A footnote admits the informal draft ran past 250,000 pages.
Instead he released the proof formalised in Lean, a programming language in which a computer checks every logical step. It comes in 14 compressed archive files, with a human-readable write-up promised later.
That changes what checking looks like. If the code builds and its final statement says exactly what the hypothesis says, the proof is, in principle, correct. MIT mathematician Andrew Sutherland told Scientific American he had compiled the result independently and called it “an historic moment”. Nobody has yet explained the mathematics from the inside, though, and that is the work specialists now face.
Who gets the $1 million prize?
The hypothesis is one of the seven Millennium Prize Problems, and the Clay Mathematics Institute offers $1 million for a solution. Nobody will collect soon. The rules require publication and years of acceptance by the field.
Why the timing raised eyebrows
The note appeared a day after Anthropic’s stock market debut, which sold its shares at $83. Alpöge gets ahead of the obvious question in the note itself. He says the breakthrough came late on Sunday night, that he messaged Dario Amodei to ask whether he could announce, and that the cynical reading of the timing “is false”.
On the same day, Alpöge and NYU’s Tristan Buckmaster posted another AI-assisted proof, on the Navier-Stokes equations, a second Millennium problem. If both survive scrutiny, 2026 is the year machines walked into the top tier of mathematics.
By Oliver Grant
Image: Slonzor / Wikimedia Commons (public domain)