AlphaGazette
Menu

Science

Anthropic mathematician claims Claude helped prove Riemann hypothesis

Levent Alpöge says a collaboration with Claude settled the 167-year-old problem and has put an argument in Lean. Anthropic has stayed silent, and no independent check is finished.

1 min read AI-generated

Levent Alpöge, an Anthropic mathematician, says he and Claude have a proof of Riemann's hypothesis, the 167-year-old question of where certain zeros of the zeta function lie. No one outside that effort has confirmed the argument, and the company has offered no comment and played no part in making it public.

In a paper this week, Alpöge wrote that he believes the joint work yields the proof. On his own site he added that he had messaged chief executive Dario Amodei. A write-up of the argument is in Lean, Scientific American reported. He thinks the method can also reach the hypothesis in its generalized form and in its extended form, and he said the reasoning holds more mathematics than he had pictured as a student. He described the find as bittersweet: the loss of a guiding light, and what he called a dawn for science.

Bernhard Riemann proposed the statement in 1859. The Clay Mathematics Institute casts it as a rule that every interesting point where zeta is zero sits on one upright line. The institute says that pattern has been tested for the first 10,000,000,000,000 such points, and that a complete proof would clarify how the primes are arranged. The question sits in the Millennium Prize set of seven problems, each tied to a prize of $1 million. A full proof had long been treated as beyond human mathematicians.

The note came in the week Anthropic floated on the Nasdaq, a listing dated November 19th, 2026. Earlier this year an OpenAI system supplied a counterexample on the planar unit-distance question, unsettling a conjecture Paul Erdős had posed. OpenAI later presented an apparent treatment of Navier-Stokes, another prize problem, then hundreds more mathematical findings. Whether those OpenAI claims hold is still unresolved.