← News

An Anthropic mathematician says he and Claude proved the Riemann hypothesis, and so far the checks agree

Levent Alpöge, a mathematician at Anthropic, said Friday that he and a non-public version of Claude found a proof of the Riemann hypothesis, and independent checks of his computer-verified proof have come back clean so far, though the final check was still running.

If this holds, an AI just helped crack the problem mathematicians have chased since 1859, the closest thing math has to a final boss. It looks real, but it isn't signed off yet, and "the computer accepted it" and "humans understand it" are two different finish lines.

On Friday, Nov. 20, Levent Alpöge posted a four-page note on his website. His words: "I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis." He says the same method also settles the Generalized Riemann Hypothesis, a wider version of the problem. The posted computer proof covers that wider claim for Dirichlet L-functions, a family of close cousins of the zeta function. The BBC reports he worked on it with a non-public version of Claude for about six weeks.

Bernhard Riemann posed the problem in 1859. It is about the zeta function, a formula tied to the prime numbers. The hypothesis says every interesting zero of that function, a point where it equals zero, sits on one straight line. If that is true, the primes are scattered about as evenly as they possibly could be. A huge amount of math already assumes it. It also carries a $1 million Millennium Prize from the Clay Mathematics Institute.

There is no ordinary paper yet. Alpöge posted the proof in Lean, a programming language in which a computer checks every step. His note says a full human write-up would be enormous and unreadable for now, and that he intends to understand it, simplify it, and write it up properly. The Lean proof spans 175,187 separate files, according to the independent check below.

Andrew Sutherland, a mathematician at MIT, rebuilt the proof independently and posted the results. As of his 4:05 p.m. CT update on Friday, every file compiled with zero errors. The main theorems rely only on Lean's three standard axioms, the basic starting rules every Lean proof is allowed to use. Two independently written proof checkers accepted the whole thing. A separate tool confirmed that the five headline theorems match the official statements, including the Riemann hypothesis as written in Mathlib, Lean's shared math library.

Two further checks were still running at that update. One is the official comparator, which checks that the code did not quietly redefine anything in the statement. The other is a full replay of the exported proof. Kevin Buzzard of Imperial College London wrote on the Lean community chat that "we still await a successful comparator run," Scientific American reports. Sutherland also lists his own caveats. The build ran without a sandbox, and a pass only shows the statement follows from the axioms as written against Mathlib's definitions.

The early verdict is confident anyway. Sutherland told Scientific American: "Honestly the chance that anything is fishy here is approximately zero." Fields Medal winner James Maynard of Oxford told Nature: "If this fully checks out, it is clearly a historic milestone in mathematics." Terence Tao told the BBC that "the digestion process will take some time, though in this case at least we have a human author to communicate with."

Not everyone is sold. Number theorist Andrew Granville of the University of Montreal told Nature, "What he released is garbled," and tied the timing to Anthropic's stock. The announcement came a day after Anthropic's Nasdaq debut. Alpöge took that on in his note: "The precise timing is horrible," and "There is an obvious cynical view, to do with the timing, but it is false." He says it all came together for him late Sunday night, and that he then messaged Anthropic CEO Dario Amodei about announcing it.

Anthropic itself has not made a statement. Nature and The Verge both say the company did not respond right away. Back in August, Anthropic said a Claude model took a real stab at the Riemann hypothesis and did not succeed, but improved a related bound, raising the share of zeta zeros known to sit on the line from 41.6% to 67.2%. Alpöge was one of the two Anthropic mathematicians who checked that work.

The picture is Levent Alpöge at the Oberwolfach Research Institute for Mathematics in Germany in 2024, during a conference on explicit methods in number theory. He is smiling in a green T-shirt, with a large round steel sculpture beside him and wooded hills behind. The credit is Marlene Ruf, Archives of the Mathematisches Forschungsinstitut Oberwolfach, CC BY-SA 2.0 DE.

In plain terms, a mathematician at Anthropic says he and a non-public version of Claude proved the Riemann hypothesis, and the independent computer checks posted so far have come back clean. The last checks were still running. Anthropic has not endorsed the claim, and the Clay prize has not been awarded.

RELATED

ONLINE…

Comments

guidelines

Loading…

Loading…

Sources