Watch the video, transcript and sources
+ 175,187 of 175,187 Lean files compile
+ con-leche accepts the 61.6 GB export in 3 min 15 s
- official lake comparator run: still going at 17:05 GMT+7Most channels are telling you Claude proved the Riemann Hypothesis. What strangers have confirmed is that about seventeen million lines of Lean compile, and that one independently written checker accepted the whole exported file in under four minutes. This is what that covers, what it leaves open, and what a builder can copy. Every status below is time-stamped, because the MIT page behind it keeps changing.
The claim. Levent Alpöge, a mathematician at Anthropic, posted a four-page note on Friday saying he believes he has, working with Claude, discovered a proof of the Riemann Hypothesis and its generalization. He released no paper, only Lean: 175,187 files. He wrote that a prose version he generated hit more than 250,000 pages, which he called slop.
Four questions. Kevin Buzzard of Imperial College published a checklist for AI-written Lean nine days earlier. Does the Lean statement match the informal one? Does the code compile? Does the code prove the theorem as stated? Does it exploit a bug in Lean's kernel? In September I told you Anthropic's Fermat formalization was thirteen million lines and that Buzzard ran out of memory trying to compile it.
Compile and kernels. Andrew Sutherland of MIT reports zero errors across all 175,187 modules and only the three standard axioms for the five headline theorems. The independently written kernels nanoda (27 minutes 42 seconds) and con-leche (3 minutes 15 seconds) accept the full export. Joachim Breitner of the Lean community called the result very thoroughly checked.
Is it the right theorem? The statement is Mathlib's own definition, whose doc comment says a proof is worth a million dollars. Seventeen million untrusted lines sit between that statement and the proof. The comparator, which I explained on November tenth, compares the trusted statement with the theorem the proof actually delivers and checks the axioms.
Where the missing check stood. As of Saturday, November 21, 17:05 GMT+7, the official lake comparator run on the full proof was still running. Its statement comparison and axiom check had passed in one run, and the slow stages replay the proof in Lean's own kernels, one thread each. con-leche's compare mode, an independent implementation of the same statement matching, accepted all five theorems. Sutherland also tried Robin's inequality, a divisor-sum statement with no zeta in it; independent checkers accept it, and he says that effectively rules out holes in the definitions behind the Mathlib statement.
The rival draft. Ravi Vakil announced a human-written draft general Riemann proof by Alessandro Masullo the same morning, reviewed for months by Brian Conrad. Eight hours later he added that people using AI models had found problems and that he makes no claim it is correct. Conrad told Scientific American it is undergoing major revision.
What a checker cannot tell you. Terence Tao wrote that digesting the proof will take some time. Scott Aaronson says mathematicians will be satisfied when they can explain the ideas themselves. Scientific American notes there are no immediate real-world applications, so nothing in your stack changes tonight.
What to copy. Read the spec before the code. Ask what compares the proof to the statement and who wrote the statement. Run a second checker written by someone else. Even build hygiene shows up: Sutherland notes the compiled files' hashes change with the build folder, since Lean can store the source path inside them.
One market footnote. Anthropic's shares fell more than eight percent on Friday, to $88.91, in a sell-off The Guardian does not tie to the proof. Neither do I, and I give no investment advice. I covered the IPO's first day yesterday.
Verdict: NEEDS REVIEW. I'd ship the checking and hold the conclusion. The compile is real, strangers' checkers agree, the official run was unfinished at the last update, and mathematicians have not yet digested the argument.
Sources
https://alpo.ge/remark.pdf
https://math.mit.edu/~drew/grh/
https://xenaproject.wordpress.com/2026/11/12/and-the-proof-is-checked-in-lean/
https://leanprover.zulipchat.com/#narrow/channel/113488-general/topic/Discussion.3A.20Alp.C3.B6ge.20Proves.20the.20Riemann.20Hypothesis
https://github.com/leanprover-community/mathlib4/blob/master/Mathlib/NumberTheory/LSeries/RiemannZeta.lean
https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/
https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/
https://scottaaronson.blog/?p=10344
https://www.scientificamerican.com/article/anthropic-mathematician-claims-to-prove-riemann-hypothesis-most-famous-problem-in-all-of-math/
https://www.scientificamerican.com/article/the-ai-riemann-hypothesis-proof-is-almost-certainly-true-independent-mathematicians-attest/
https://www.scientificamerican.com/article/dueling-claims-of-riemann-hypothesis-proofs-emerge/
https://www.theguardian.com/business/2026/nov/20/global-stock-markets-slump-anthropic-ipo-ai-sell-off
https://news.ycombinator.com/item?id=50574968
And that's the diff for today. I'm Niko from Axrisi. Merge responsibly.
YouTube · Newsletter · thedailydiff.dev · forward this to whoever trusts a green check without asking what it compared.
Credits: Portrait of Terence Tao at IPAM's Fireside Chat 2026, Institute for Pure & Applied Mathematics, CC BY 4.0 (licence, Commons file); cropped, background removed; identifies Tao only, no endorsement, quote or reaction (he wrote about the claim and did not say it was proved). Portrait of Dario Amodei at TechCrunch Disrupt 2023, by TechCrunch, CC BY 2.0 (licence, Commons file); background removed, cropped, resized; identifies Anthropic only. Two-second silent quotation of the "Double checking" corgi GIF via GIPHY; the owner keeps all rights and nobody in this story reacted. "Is This A Pigeon" meme template via imgflip.

