Morning Edition · Sunday, November 22, 2026Published at 1:11 AM EST · New York
The proof is a 17-million-line Lean library from Anthropic mathematician Levent Alpöge, and it compiles against Mathlib's own statement of the conjecture. No human-readable paper exists, and the Clay Mathematics Institute still lists the problem as unsolved.

The Lean Focused Research Organization (FRO) and the Mathlib Initiative issued a joint statement on Saturday: "We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct." AI Post and Unite.AI both reported it. Levent Alpöge, a mathematician employed by Anthropic, released the proof on November 20. It is a Lean library of about 17 million lines that he produced with Claude, according to Kingy AI.
The checking covers three points. First, the theorem proved is Mathlib's existing definition of the Riemann Hypothesis, not a statement Alpöge wrote himself. Two tools, comparator and con-leche, reportedly confirmed that match. Second, Andrew Sutherland, a mathematician at the Massachusetts Institute of Technology (MIT), reported that all 175,187 modules compiled with zero errors on Lean and Mathlib v4.33.1. He said he ran those checks with Claude Code. Third, Lean's kernel and several other checkers reportedly accepted the exported proof term (Kingy AI). If all three hold, a reader only has to trust the Lean kernel, the standard axioms, and the claim that Mathlib's definition states the conjecture correctly. No referee has to read 17 million lines.
Human mathematicians have not yet independently confirmed the proof through traditional peer review, and no paper describing the argument exists yet. Terence Tao's first reading is that the strategy draws on the Langlands program and arithmetic geometry, and he expects mathematicians to need time to see how the parts fit together (OfficeChai). The Clay Mathematics Institute's page still listed the problem as unsolved on November 21. Its prize rules require publication in a qualifying outlet, a two-year wait, and general acceptance by mathematicians (Nacionale). Alpöge's note also claims the Generalized Riemann Hypothesis, but the Saturday statement does not mention the generalized version.
This result follows a sequence of formally checked machine output. An earlier arXiv paper reported, and formalized in Lean, that more than 83.9 percent of zeta zeros are distinct and more than 67.35 percent are simple and on the critical line. On November 10, Google DeepMind's AlphaProof Code reportedly wrote about 6.5 million lines of Lean proof about the compiled machine code of Signal's libsignal library in four weeks (Xenospectrum). Machine-written proofs now run to millions of lines. At that size no human can review them line by line, so a passing kernel check has become the practical standard for accepting them.
The person making the claim has a stake in it. Alpöge works for Anthropic, which listed on Nasdaq on November 19, and a verified result supports its pitch to public investors. That is why it matters that the checkers are independent of Anthropic: the Lean FRO, Mathlib's maintainers, and Sutherland. Anthropic shares closed Friday at $88.91, before the statement came out. Monday's session will be the first market reaction.
Part of a tracked trend
Formal Verification Becomes AI's Trust Layer
As model-generated proofs and code grow past what humans can review, machine checking in Lean and similar systems becomes the accepted way to verify AI output in mathematics and security-critical software, and labs will increasingly ship formal certificates instead of benchmark claims.
Start a discussion in Townsquare.
More from this edition
Anthropic gains most because it is newly listed and can now show public investors a result that Lean's maintainers have certified, and labs with mature Lean tooling gain because a kernel check is becoming the accepted standard for proofs too long for humans to review.
The certification confirms only that the code compiles and that its theorem matches Mathlib's formal statement of the conjecture, and Scientific American reported skepticism from mathematicians alongside a competing human-written proof claim, so whether the mathematical community accepts the argument depends on a human-readable write-up that Alpöge has said he intends to produce but has not yet released.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
A kernel-checked proof turns a lab's math claim into a file anyone can rerun, which benefits labs with strong Lean tooling, such as Anthropic and Google DeepMind. The desk expects other labs to ship Lean certificates with their announcements within months, unless a flaw surfaces in Mathlib's statement of the conjecture.
What to watch
Observations to monitor, not financial advice.
Synthesized from: Polylog editors · Unite.AI · Kingy AI · OfficeChai · Nacionale · arXiv · Xenospectrum
Comments
0No comments yet.