The Lean and Mathlib communities say they have verified a Claude-assisted proof of the Riemann hypothesis, a 167-year-old problem about the distribution of prime numbers. Their November 21 joint statement endorses the correctness of both the mathematical statement and Levent Alpöge's computer-checked proof.

That is a new step beyond Alpöge's announcement that he believed he had found a proof. The Clay Mathematics Institute's Riemann hypothesis page still labels the problem unsolved, and its prize process has separate publication and waiting requirements.

What the proof claims

Archival portrait of Bernhard Riemann
Bernhard Riemann in 1863. The photographer is unidentified.Wikimedia Commons, public domain

Alpöge, a mathematician working at Anthropic, described his collaboration with Claude in a four-page note. He released the argument as Lean code, a formal language for mathematics, rather than a conventional paper explaining the whole proof.

In the note, he says he guided the work toward a setting that combined two mathematical ideas, then credits Claude with much of the later progress. He also says he intends to understand, simplify and explain the long argument more fully. His account describes a collaboration with a mathematician directing the work.

The question dates to Bernhard Riemann's 1859 paper. Prime numbers, including 2, 3, 5 and 7, cannot be written as products of two smaller positive whole numbers. The prime number theorem describes their average distribution. The Riemann hypothesis concerns how far their count can depart from that average.

Its familiar technical statement says the nontrivial zeros of the Riemann zeta function all have real part one-half. Clay's overview says the first 10 trillion relevant zeros have been checked. A proof must cover every case, including the infinitely many beyond any finite computation.

What the independent checks establish

The community statement describes a library of about 17 million lines of code. Its verification addresses several different questions: whether the code compiles, whether it proves the intended theorem, and whether the resulting proof passes logical checking.

Matching the intended statement requires its own check. A program can successfully prove a statement that differs from the question readers think it answers. The community says it isolated Mathlib's definition of the Riemann hypothesis and used two tools to compare that statement with the theorem in Alpöge's library.

The statement also says multiple proof checkers accepted an exported version of the argument. Those checks operated on an inert export, outside the reach of hypothetical executable code in the original source.

MIT mathematician Andrew Sutherland's verification record supplies more detail. It reports that all 175,187 modules compiled and that the headline theorems used only Lean's three standard axioms. It also records accepted checks by independently written kernels.

That record distinguishes a checker that rejects a proof from an older checker that cannot process a feature of its format. It lists some runs as still in progress. The separate community statement gives the broader endorsement; individual status logs retain their own update times.

Sutherland also limits the generalized result's scope to Dirichlet L-functions, across all moduli and characters. His record does not claim verification for Dedekind zeta functions, Hecke L-functions or automorphic L-functions generally.

A separate outline by mathematician Siniša Slijepčević, dated November 21, describes a reduction to bounds on weighted prime-counting functions. It says those bounds come from mathematical objects called Hermitian lattices, with other estimates controlling the losses in the inequalities. Its opening theorem rules out zeros to the right of the one-half line for Dirichlet L-functions. Slijepčević says the AI-assisted outline covers only part of the development and does not verify it.

Why the prize process takes longer

Opening page of the historical report containing Riemann’s work on prime numbers
A scan of the report containing Riemann’s 1859 work on prime numbers, published in Berlin in 1860.Bernhard Riemann / Wikimedia Commons, public domain

A community verification announcement does not complete Clay's award process. Under the institute's published rules, a proposed solution must appear in a qualifying outlet, at least two years must pass after publication, and the work must gain general acceptance in the global mathematics community before Clay will consider it.

Clay also says it does not accept proposed solutions submitted directly to the institute. Posting code and receiving a formal-verification endorsement therefore do not, by themselves, meet the full set of conditions for prize consideration.