FULL STORY

Riemann Hypothesis 'AI Proof': From Hype to Doubt

A claimed Lean-formalized, AI-assisted proof of the Riemann Hypothesis by Levent Alpöge went viral before mathematicians raised a wave of technical objections, with some calling the proof invalid and the saga mired in ongoing controversy.

2026-11-21 ~ 2026-11-23 · 5 episodes · 14 posts

Episode 1 · Mathematician Claims Claude-Assisted Proof of Riemann Hypothesis, Verified in Lean (2026-11-21, 2 posts)

Mathematician Levent Alpöge claims to have proved the Riemann hypothesis using Anthropic's Claude, producing a 25-million-line, roughly 250,000-page Lean-formalized proof verified by Lean but too long for anyone to read. He argues AI is ending the era when mathematics required trusting no one.

Episode 2 · Doubts Mount Over Alpoge's Formalized Riemann Hypothesis Proof (2026-11-21, 2 posts)

Mathematicians continue to challenge Alpoge's Lean-formalized proof of the Riemann Hypothesis, arguing it passes type checking only because the geometric and analytic sides are definitionally identical, making the theorem vacuous.

Episode 3 · Mathematician Claims AI-Assisted Riemann Hypothesis Proof Invalid Despite Lean Verification (2026-11-22, 2 posts)

Mathematician Jenny Lorraine Nielsen argues that the AI-assisted proof of the Riemann Hypothesis is invalid despite passing Lean formal verification, casting doubt on the reliability of AI-driven mathematical proofs.

Episode 4 · Viral "Riemann Hypothesis Proved" Claim Is an AI Community Prank (2026-11-22, 3 posts)

A viral "joint statement" claiming Lean FRO and Mathlib verified a 17-million-line Lean proof of the Riemann hypothesis by Levent Alpöge, along with a parody ACM announcement crediting Claude, is actually an elaborate AI-community prank spawning memes like "Claude's lemma."

Episode 5 · Lean Riemann Proof Officially Verified but Trust Debate Persists (2026-11-23, 5 posts)

Alpöge released an AI-assisted formalized proof of the Riemann Hypothesis as a Lean library of roughly 17 million lines (about 250,000 pages), and debate over its validity and mathematical value evolved rapidly within days. Official verification has now landed, but the trust controversy has not subsided.

Confirmed

  • The Lean and Mathlib communities issued a joint statement: all 175,187 modules compiled successfully, and multiple independent kernels verified the roughly 17 million lines of code with consistent results.
  • According to @MickeySteamboat, Lean's actual maintainers officially confirmed on November 21 that the proof corresponds to the standard mathematical formulation of the Riemann Hypothesis with no extra assumptions smuggled in; @QualiaQuanta's teardown thread likewise analyzes it on the premise that it passes kernel type-checking.
  • Multiple independent verifiers produced consistent results matching the standard statement of the Riemann Hypothesis.

Unconfirmed

  • @QualiaQuanta pointed out that the technical reason the proof passes kernel type-checking is that the geometry side and the analysis side are defined as the same object, citing definitions like anyGeometry := fun => … where the geometric predicate appears to be trivially True; others argue that once any part of a 250,000-page "slop" requires "trust," its correctness and mathematical value become fatally compromised. The tension between these objections and the official conclusion that "formal verification holds" has yet to be fully resolved.

Why it matters

  • If the formalized proof is ultimately fully accepted by the mathematical community, it would be a landmark for AI-assisted mathematics; even so, the "trust model," readability, and mathematical value of ultra-large machine-generated proofs have become issues the formal verification community must confront head-on.

Timeline

  • Alpöge releases the roughly 17-million-line Lean proof library.
  • "Slop" criticism and pushback from supporters erupt on X.
  • November 21: Lean maintainers officially confirm the proof; multiple independent verifications agree.
  • Around November 23: users (@QualiaQuanta and others) continue dissecting the proof's internal definitions.