Latest
AI Affairs, home

Saturday 21 November 2026

Technology

Anthropic mathematician claims Claude helped prove the Riemann hypothesis

Levent Alpöge has released a version of his claimed proof in Lean, but mathematicians still need to check whether the code compiles and the argument holds.

A young man in a red beanie writes equations on a chalkboard.
Photo: Yan Krukau via Pexels

On 20 November 2026, Levent Alpöge, a mathematician at Anthropic, said in a statement on his website that he had discovered a proof of the Riemann hypothesis while working with Claude, Scientific American reported. Alpöge also posted a version in Lean for mathematicians to check mechanically, New Scientist reported.

Key points

  • Alpöge says he found a proof of the Riemann hypothesis with Claude.
  • He released a Lean version but says a human-readable version remains unpublished.
  • New Scientist says humans need to check the code and mathematics; Kevin Buzzard of Imperial College London questioned whether the code compiles.

Riemann’s 1859 problem

Bernhard Riemann proposed the hypothesis in 1859. It concerns a pattern underlying the distribution of prime numbers and is among the Clay Mathematics Institute’s Millennium Prize Problems, which carry a $1 million prize for an accepted answer. Alpöge says his argument also proves the Generalized Riemann Hypothesis, a broader version of the original conjecture.

In July, Alpöge announced that he had used Claude to produce a 216-character counterexample to the Jacobian conjecture. He appeared to suggest that he held back his Riemann announcement until after Anthropic’s stock-market listing on 19 November, New Scientist reported. Alpöge said he had spoken with Anthropic chief executive Dario Amodei and found that “there was a constraint to respect”.

Alpöge’s Lean code faces inspection

Lean expresses a mathematical argument as small logical steps that software can check, much as a calculation is checked one line at a time. Alpöge released his claimed proof in that form rather than publishing a human-readable account, which he says ran to more than 250,000 pages.

New Scientist said humans would need to inspect both the code and the mathematics. Kevin Buzzard, a mathematician at Imperial College London, turned first to whether the code compiles. “Let’s see if it compiles first,” he said.

Sources

Topics: Agents, Foundation models

What should we cover? Tell us

The weekly briefing

The AI Affairs weekly briefing: the week's 10 most important AI stories in business, government and markets, every Monday. We'll email you a link to confirm. Every email has a one-click unsubscribe. Privacy policy