20 November 2026 · Levent Alpöge / Anthropic; Alessandro Masullo / Stanford
An Anthropic mathematician claims a proof of the Riemann hypothesis with Claude; a Stanford draft claims the generalized version
On 20 November Levent Alpöge, a mathematician at Anthropic, published a four-page note saying he believes he has proved the Riemann hypothesis working with Claude, and that the method also covers the generalized version. He released the argument only as Lean code, a format in which a computer checks every step. No outside expert has confirmed it yet. The same day a second claim appeared, to the generalized hypothesis, in a human-written draft from Stanford.
CapabilityEarly signalFrom someone in a position to know, or a post many people are reading. Not confirmed by a second source yet. I follow it up.
What happened
Alpöge writes on his website that he believes he has discovered a proof working with Claude, that the method also resolves the generalized Riemann hypothesis, and that he expects it to reach the extended version too, given more time. He says the arguments are formalized without assumptions. He did not post a conventional paper: the argument is very long, and he writes that posting a huge unreadable document would be inappropriate, so he presents only the Lean code, in fourteen archive parts.
He also writes that he has little mastery of automorphic forms on higher-rank groups, and that Claude moved far ahead of him near the end while he steered the overall approach. So the technical heart of the argument is, by his own account, mostly Claude's work.
What has been checked so far is little. Scientific American writes that experts still have to confirm the statement is coded correctly in Lean and that the proof holds. One write-up of the first public checks says an independent copy of the files matched the release and that five independently written theorem statements compiled against Mathlib, Lean's mathematics library, while an end-to-end verdict and a separate replay by the Lean kernel were still pending. The Clay Mathematics Institute still lists the hypothesis as unsolved.
Alpöge works at Anthropic and the model is his employer's. In the note he calls the timing awful, says he messaged Dario Amodei late one Sunday night and asked about announcing, and says the leadership and the few mathematicians he consulted held similar views on what to do. He acknowledges a cynical reading that has to do with the timing and calls it false. The note does not say what the timing refers to. Anthropic's shares began trading the day before.
The second claim is a draft by Alessandro Masullo, a Stanford-trained mathematician, posted on his blog by Ravi Vakil, the president of the American Mathematical Society. It claims the generalized hypothesis 'by an entirely human route', after twelve years of work, as a written paper and not as Lean code. It is unfinished and not peer reviewed. Stanford's Brian Conrad, who has read it for months, told Scientific American it is undergoing major revision. Vakil writes that the plan was to bring Masullo to Stanford for a year to polish it first, and that it now seemed best to release it as it stands.
What came before
Alpöge's note includes a rough timeline that Claude produced, running from early October to mid-November, and he warns that parts of it are summary he has not yet checked and that the Lean should be what counts. For how the company sees itself, see the view from inside; the note does not discuss Anthropic's strategy.
What I do not know
Whether the Lean statements match the real Riemann hypothesis, and whether the whole archive compiles and holds, has not been confirmed by anyone outside Anthropic in the sources read.
One outlet counts 175,187 Lean files in the archive and another headline says it was verified by MIT kernel checks. Neither fits the other sources, which say verification is pending, so neither is used here.
One aggregator reports that Vakil later said problems had been found in Masullo's draft with the help of AI. That was not found in any other source.
The exact page address of Alpöge's note has not been confirmed. The note was read as a PDF; its footnote places the archive parts at alpo.ge.
Read next
Sources
- Levent Alpöge: 'Remark' (four-page note)primary · main source
Read as a PDF. The page address is the site's front page, not confirmed as the note's own address. Source for what Alpöge claims, his credit to Claude, the timing remarks and the fourteen Lean archive parts.
- Nature: Mathematician says he and AI solved Riemann hypothesispress · not read end to end yet
Seen through a search snippet. Confirms the claim, who made it and that it is Anthropic's.
- Scientific American: Anthropic mathematician claims to prove Riemann hypothesispress · not read end to end yet
Seen through a search snippet. Source for experts still needing to verify the Lean statement and the proof.
- IBM Think: Anthropic mathematician claims AI helped solve a 167-year-old math mysterypress · not read end to end yet
Seen through a search snippet. Confirms that the argument was released as Lean and not as a conventional paper.
- Kingy AI: Levent Alpöge claims a Riemann hypothesis proof with Claudepress · not read end to end yet
Seen through a search snippet. Source for the first public checks and for the Clay Institute still listing the problem as unsolved. A small blog; treated as one report.
- DataCamp: Two proofs of the Riemann hypothesis dropped on the same daypress · not read end to end yet
Seen through a search snippet. Source for the two claims landing the same day. Its quick answer is that neither is confirmed.
- Yahoo Tech (TechCrunch): AI claimed a solution to the 167 year-old Riemann Hypothesis, and a human did, toopress · not read end to end yet
Seen through a search snippet. Source for Masullo's twelve years of work.
- TokenPost: Stanford professor releases human-written general Riemann hypothesis draftpress · not read end to end yet
Seen through a search snippet. Source for Vakil releasing the draft ahead of a planned year of editing. Vakil's own blog post has not been found.
- Nacionale: Why Masullo's 'all human' proof draft matterspress · not read end to end yet
Seen through a search snippet. Source for Conrad's remark and for the draft being unfinished and not peer reviewed. An aggregator that cites TechCrunch and Scientific American.