Prove the Riemann Hypothesis and formalize it in Lean

Anthropic mathematician Levent Alpöge says he and Claude found a proof of the Riemann Hypothesis and wrote it out in Lean. Within a day independent mathematicians had compiled the Lean proof.

Done withClaude· Nov 20, 2026 · 5 accounts

What they did

Levent Alpöge, a mathematician who works at Anthropic, worked with Claude on the Riemann Hypothesis, the 1859 conjecture about where the zeros of the zeta function lie and one of the seven Millennium Prize problems. He posted a short note on his website saying the two of them found a proof, that the method also covers the Generalized Riemann Hypothesis, and that he and Claude formalized the argument in Lean, the proof checker, with no added assumptions. The proof is published as Lean code rather than a conventional paper. The Verge reports he had earlier said he settled the Jacobian conjecture with Claude, and had been working on Navier-Stokes with an NYU professor.

How it went

At announcement the result was unverified, and the Clay Mathematics Institute requires publication and two years of acceptance before it counts. By the next day Scientific American reported that independent researchers, including MIT mathematician Andrew Sutherland, had compiled the Lean proof themselves and the kernel rechecked it using only the three standard axioms; Sutherland put the chance of anything being wrong near zero. Kevin Buzzard said the remaining step was a comparator run to confirm the code does not redefine any symbol in the statement and compiles under several independent kernels. Alpöge says he wants to simplify and explain the argument himself.

Worth knowing

A Lean proof shifts the question from reading hundreds of pages to checking that the code compiles and that the final statement really is the standard one in Mathlib. Formal proofs can still hide an extra assumption or a mismatched definition, which is what reviewers will look for. The harness Claude ran in was not named. The Neuron reports the Lean files came in 14 archive parts and that Claude found a workaround built from a long chain of theta lifts; Kevin Buzzard's first reaction was to see whether it compiles.

Try it

With Claude· 3 parts to fill in

Work with me on [open problem or conjecture] in [field]. Start by stating it precisely as it appears in [Mathlib or the reference I give]. Propose lines of attack, test each against known special cases, and tell me plainly which ones fail. For any approach we agree on, write the argument out in Lean with no new axioms or sorry, and keep the final theorem identical to the standard statement. After each milestone, stop and show me what compiles, what is still assumed, and where you are unsure, before going further.

Square brackets are the parts to fill in. The prompt is ours; the story is theirs.

Where they said it

Related