FULL STORY

Claude's Claimed Riemann Proof: From Hype to Skepticism

Anthropic mathematician Levent Alpöge claimed a 250,000-page Lean 4 proof of the Riemann Hypothesis co-authored with Claude, touting 99% verification. The claim went viral before facing skepticism, with Polymarket pricing proof odds at just 6%.

2026-11-20 ~ 2026-11-21 · 5 episodes · 213 posts

Episode 1 · Anthropic Mathematician Claims Riemann Proof with Claude; Masullo's Human-Only GRH Draft Emerges (2026-11-20, 191 posts)

Breaking news in mathematics: during an AMS executive committee meeting, Stanford math professor Ravi Vakil urgently posted an announcement that mathematician Alessandro Masullo has produced a draft proof of the Generalized Riemann Hypothesis (GRH), roughly 940 pages long, stressing that the proof is "entirely human"—100% independently completed with no AI involvement. The preprint draft has been released and is now under review.

Confirmed

  • The source is Stanford professor Ravi Vakil's announcement at the AMS meeting and his blog post, amplified by several mathematicians (e.g., Alex Kontorovich).
  • The proof draft is roughly 940 pages, authored by Alessandro Masullo, and claims no AI involvement whatsoever.
  • Stanford professor Brian Conrad has spent months carefully reviewing the draft and provided extensive feedback; according to Vakil, things look good so far.
  • Talks related to Masullo's work were originally planned (details per Vakil's original post).

Not Yet Confirmed

  • Whether the proof is actually correct remains open. Conrad's review is still ongoing, and "looks good so far" reflects only a first impression—results at the level of GRH typically require lengthy peer verification before any conclusion.

Why It Matters

  • The Generalized Riemann Hypothesis is one of the most central open problems in number theory and mathematics as a whole; a proof would have profound consequences.
  • Against the backdrop of increasingly common AI-assisted research, a massive proof explicitly declared "purely human, no AI involved" is itself newsworthy.
  • The formal announcement by a senior mathematician like Vakil, plus Conrad's months of review, makes the community take this far more seriously than typical crank claims of "proving Riemann"—though the final verdict must await the full review.

171 more related posts →

Episode 2 · Claim of Claude Proving Riemann Hypothesis Debunked (2026-11-21, 2 posts)

An Anthropic mathematician claimed Claude proved the 167-year-old Riemann Hypothesis, but the 250,000-page output was quickly debunked as AI slop starting with a tautology, drawing mockery from researchers.

Episode 3 · Claimed 250,000-page Riemann Hypothesis proof by Levent and Claude sparks memes and skepticism (2026-11-21, 13 posts)

On November 21, word spread fast across math/AI circles on X that Levent and Claude had completed a full proof of the Riemann Hypothesis—reportedly over 250,000 pages long, so vast that even the human "author" can't fully parse it. Multiple claims of related proof releases surfaced the same day, sparking massive discussion and memes.

Confirmed

  • The circulating claim states the proof was done by Levent in collaboration with Claude and exceeds 250,000 pages (per ArthurMacwaters; MattPirkowski also said over 250,000 pages, while nrehiew called it a "million-page proof")
  • The event triggered wide discussion and meme floods: schaferkar l self-mocked with the classic "they don't know RH has fallen yet" format; eigenron's "imagine being the Riemann Hypothesis and getting taken down by a bunch of matrix multiplications and nonlinear activations" became the most viral joke of the cycle
  • The proof itself awaits verification; nrehiew noted nobody will actually read it, and the only credible "signal" is whether the Lean formalization compiles (he thinks it of course will)

Unconfirmed

  • The proof's authenticity and correctness lack independent verification; all posts are commentary and secondhand accounts with no primary confirmation
  • Posts disagree on the length (250,000 vs. a million pages); the exact figure needs checking

Why it matters

  • theojaffee and gleech relayed a sober skepticism: perhaps the takeaway is just "the Riemann Hypothesis is easier than we thought" rather than "AI is about to massively surpass humans at nearly everything"—we don't actually know how hard math problems are; gleech added that AI will soon scale this ability to "chain known points"
  • MattPirkowski used this to advance a "re-enchantment of computation" thesis: the scalability of computational verification is outstripping human comprehension; nrehiew saw the no-one-can-read-it situation as confirming the trend. ArthurMacspaces/ArthurMacwaters offered a mental model: many future scientific/mathematical discoveries will arrive in "alien machine language," with humans reading only conclusions, unable to go through the arguments

Episode 4 · Live Lean Verification of Riemann Hypothesis Proof Reaches 99% (2026-11-21, 5 posts)

Mathematician Levent Alpöge has uploaded a Lean 4 formalization claiming a complete, unconditional proof of the (generalized) Riemann Hypothesis, done in collaboration with Anthropic's Claude. The community immediately began public verification: one user set up a public website livestreaming the process on a server with a 128-core CPU and 244GB of RAM, downloading all 14 archives totaling 228.2MB across 17 files (material truncated); as of November 21, verification had reached 99%, with the livestream showing compilation expected to take about 3 hours.

Confirmed

  • The proof was uploaded by mathematician Levent Alpöge, claimed as a complete unconditional proof of the (generalized) Riemann Hypothesis, submitted as a Lean formalization
  • altryne opened a live stream tracking the Lean 4 verification and provided a backup link after the original site had access issues, crediting @JunyanXu
  • The live verification runs on a 128-core CPU, 244GB RAM server, involving 14 archives totaling 228.2MB
  • Per m1, verification progress has reached 99%
  • Alex Kontorovich walked back his earlier "re-running the code is silly" remark

Unconfirmed

  • m4 noted the proof's original author was still being identified at the time; later posts point to Levent Alpöge
  • Even if the Lean verification fully passes, whether the proof is accepted by the math community remains to be seen

Why it matters

  • The Riemann Hypothesis is one of mathematics' most famous open problems; passing formal verification would be hugely significant
  • AI (Claude) playing a deep role in formalizing a proof of a major conjecture showcases AI-assisted mathematical research
  • Mathematician Alex Kontorovich noted it's important to re-run machine verification across different Lean implementations; what's truly pointless is watching the compile live—it will pass. For human mathematicians, the value lies in reading and understanding the proof itself

Episode 5 · Polymarket Odds: Only 6% Chance Riemann Hypothesis Proven by 2026 (2026-11-21, 2 posts)

After an Anthropic researcher released a Claude-assisted claimed proof of the Riemann Hypothesis, Polymarket opened a market pricing only a 6% chance that the Clay Mathematics Institute announces a proof by end of 2026, and 14% by 2027, with about $104K in volume.