AI Model Comparison

Model profile & tracker

The Riemann Hypothesis AI Claim: What's Actually Known

The 20 November 2026 Riemann hypothesis claim by Larry Alpöge and Claude: what was claimed, verification status, and the Clay Millennium prize rules.

Verified 2026-11-22

On 20 November 2026, Larry Alpöge — a mathematician at Anthropic — posted a claimed proof of the Riemann hypothesis, co-credited to the AI system Claude. Within hours it was the most-read mathematics event of the decade. Whether it holds is, as of 22 November, an open question. This page tracks the claim's status and will be updated as verification proceeds.

What was claimed

The preprint claims a proof of the Riemann hypothesis: the 1859 conjecture that all non-trivial zeros of the Riemann zeta function have real part ½. It is one of the seven Clay Millennium Problems, carrying a $1,000,000 prize, and the foundation of most of what is believed about the distribution of prime numbers.

What makes this claim unusual is the attribution. The work is credited jointly to Alpöge and to Claude, with the human author describing his role as guiding the system toward the key constructions. The preprint runs to roughly 175,000 lines of Lean 4 proof-assistant code across some 250,000 pages of supporting material — machine-checkable in principle, though nobody has independently verified the full development yet.

The verification status, as of 22 November 2026

  • The claim is unverified. No independent group has confirmed the proof. The Clay Mathematics Institute requires publication in a refereed journal plus a two-year waiting period before any prize is awarded.
  • The Lean development is not yet externally checked end-to-end. Compilation of the full development is the acid test; the preprint has not cleared it publicly.
  • Mathematicians are cautiously engaged rather than dismissive. James Maynard called it "a historic milestone" — with the caveat that the proof still needs to survive verification. The same pattern held for the September claim on Navier-Stokes: serious attention, no acceptance yet.
  • One known gap in the generalised claim. The paper also claims the generalised Riemann hypothesis (GRH) over number fields; early readers report a gap in that portion. The classical statement — the one worth the Millennium prize — is what the verification effort is concentrating on.

Why it matters, in three lines

The Riemann hypothesis controls the error term in the prime-counting function. A proof would collapse a century of conditional results in analytic number theory and cryptography into unconditional ones. And if the proof is machine-checkable, it becomes the largest mathematical argument ever verified by software rather than by peer review alone.

What the number says

Verification of zeros continues in parallel with the proof attempt: the latest computations confirm the hypothesis for the first 67.25% of the non-trivial zeros on the critical line. That is evidence, not proof — but it is the strongest numerical evidence ever assembled, and it is what any correct proof must be consistent with.

The bigger picture

This is the third major AI-assisted mathematics claim of the season: after the Navier-Stokes work in September, a claimed Riemann proof lands within two months. The pattern — human mathematician guiding a frontier model through a proof search, with the output written in Lean — is becoming the standard shape of these attempts. This site tracks AI models, not mathematics, but the models making these attempts are in the comparison table: Claude Fable 5.2 among them, and the frontier releases of November 2026 — Gemini 4 Neon, Qwen 4 Max — are the generation that follows.

FAQ

Has the Riemann hypothesis been solved? As of 22 November 2026: claimed, not solved. Larry Alpöge and Claude posted a proof on 20 November 2026; verification is ongoing, and the Clay Institute's two-year rule applies regardless.

Who is Larry Alpöge? A mathematician working at Anthropic. He is the human author of the Riemann claim, with Claude credited as co-author.

What does the Riemann hypothesis say? That every non-trivial zero of the Riemann zeta function has real part exactly ½. It has been open since 1859.

Why is the Lean code important? A machine-checked proof cannot contain hidden gaps. If the development compiles, the proof is correct by construction — independent of whether reviewers can read all 250,000 pages.

Could an AI really prove it? The claim is that a human guided the model to the proof, not that the model produced it unsupervised. That distinction is exactly what mathematicians are now arguing about.