
The Anthropic mathematician’s claimed proofs of the original and generalized hypotheses have not been independently verified or publicly accepted by the Clay Mathematics Institute. A separate code check is underway.
Anthropic mathematician Levent Alpöge says he and Claude, the company’s model, discovered and formalized proofs of the Riemann hypothesis and its generalized version without extra assumptions. The claim has not been independently verified or publicly accepted by the Clay Mathematics Institute, which administers the Millennium Prize.
Bernhard Riemann proposed the original hypothesis in 1859. It concerns a function linked to the distribution of prime numbers and is one of seven Millennium Prize Problems; only the Poincaré conjecture has been officially solved.
A note and code appeared on Alpöge’s website on 20 November 2026. The note credits Claude as a collaborator and says a human-readable version will come later. The code is written in Lean, a programming language that allows a computer to check the logical steps of a proof.
Claude and I have also formalized the arguments without assumptions.
Alpöge says he expects the method to extend to the extended Riemann hypothesis with more time; he does not say he has proved that version. The Verge says Anthropic has not commented on or involved itself in his announcement.
The original hypothesis says that every relevant, or nontrivial, zero of the Riemann zeta function has a real part of one-half. A zero is a value for which the function’s output is zero; its real part is one component of a complex number. The generalized hypothesis makes the same assertion about zeros of a family of related functions called Dirichlet L-functions.

A verified proof would establish results that mathematicians have proved on the assumption that the hypotheses hold, and would sharpen bounds on estimates of how primes are distributed.
In August, Anthropic said an unreleased research version of Claude had improved a lower bound on the proportion of zeta zeros on the relevant line from 41.6% to 67.2%. The company said that result did not prove the hypothesis.
A separate verification repository says its copy matches the published archive, its test statements compile against the Mathlib library alone, and a scan found none of the listed prohibited constructs. Its Comparator check of whether the statements match, which axioms are used and whether the proof can be replayed is still running.
The checks must establish that the completed code proves the intended statements and relies only on standard axioms, or starting assumptions. The repository says even a passing check would still depend on components including Lean’s core checking system and Mathlib’s definitions.
The institute offers $1 million for each Millennium Prize Problem. Before it will consider a solution, the work must be published in a qualifying outlet, at least two years must pass after publication, and the global mathematics community must generally accept it. The institute does not accept direct submissions.
Further checks matched five theorem statements and found only standard starting assumptions, but mathematical review of the argument continues.
3 min