scientificamerican.com web signal

Anthropic's Alpöge Posts Riemann Hypothesis Proof With Claude

7 sources tracking this story
Anthropic ai-business

TL;DR

  • The Lean FRO and Mathlib Initiative certified the proof correct on November 21 using independent checkers con-leche and con-ron, each with formal soundness proofs relative to set theory.
  • The proof covers the Generalized Riemann Hypothesis as well, extending the scope beyond what initial reports described.
  • Alpöge produced three Millennium-class results in five months: Jacobian conjecture in July, Hopf's problem in August, and Riemann Hypothesis in November.

Levent Alpöge, a mathematician employed at Anthropic, said on Friday he had proven the Riemann hypothesis, the problem Bernhard Riemann first proposed in 1859 and one of the few carrying a million-dollar bounty from the Clay Mathematics Institute. Scientific American reported the claim as an AI-generated solution of the 167-year-old problem.

He did it with Claude.

"I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," Alpöge wrote on his personal website. In a more personal register he added: "It is natural to feel that this is an end, in my case a loss of a principal guiding light."

The proof is posted as code in Lean, the programming language that lets a machine verify each step of a mathematical argument. Hector Pasten, a mathematician at the Pontifical Catholic University of Chile, told the magazine that if the argument holds, it "completes a dictionary between different branches of math where the connections were only partial. Something like a Rosetta Stone."

No outside mathematician has confirmed the work. Anthropic has issued no statement. Scientific American notes that Lean checking should not take long, but experts still need to verify both that the hypothesis has been properly coded and that the underlying argument is correct.

What others are reporting

Coverage cluster as of 24h after publish

  1. Nature Read →

    Features Fields Medal winner James Maynard's conditional endorsement alongside Andrew Granville's skepticism about garbled documentation and the suspicious proximity to Anthropic's IPO drop.

    If this fully checks out, it is clearly a historic milestone in mathematics.
  2. Lean Prover Community Read →

    Official joint statement from the Lean FRO and Mathlib Initiative detailing the four criteria met, with confirmation by multiple independent proof checkers beyond Lean's own kernel.

    We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.
  3. IBM Think Read →

    Foregrounds the comprehension gap: even after machine verification, experts warn that understanding the proof's mathematical reasoning could take months or years.

  4. Crypto Briefing Read →

    Places the Riemann claim inside a 2026 productivity streak, framing Jacobian in July, Hopf's problem in August, and Riemann in November as evidence of an accelerating AI-math cadence.

    A machine can confirm that a proof is valid. Humans still have to confirm that the proof is of the right thing.
  5. Unite.AI Read →

    Highlights the methodological transparency gap versus Alpöge's August zeta-bound work, which underwent review by Brian Conrey and Dan Goldston before release.

    The release therefore pairs a formalization presented as complete with limited human-readable explanation.
  6. CASRAI Read →

    Focuses on institutional fallout: what universities must now build in archival systems, peer review frameworks, and authorship attribution guidelines for AI-assisted mathematics.

    The announcement marks the first instance of a Millennium Prize Problem being claimed through a fully compiled formal proof assistant codebase rather than a conventional academic manuscript.

Shared on Bluesky by 9 AI experts (top 5 by trust)