Claude-aided Riemann Hypothesis proof claimed in Lean
Anthropic mathematician Levent Alpöge says he and Claude proved the Generalized Riemann Hypothesis, and posted the proof only as Lean code, with no human-readable paper yet. The checker matters more than the hype: if the Lean holds, the 167-year-old question is settled before anyone fully understands why.
- BY
- TopFive Desk
- FILED
- READ
- 4 min
- SRC
- TechCrunch ↗
Either the most famous open problem in mathematics just fell to a person working with an AI, or the world is about to learn how a formal proof checker can be fooled.
Both outcomes matter. Only one looks likely.
Levent Alpöge, a mathematician at Anthropic, posted a short note on Nov. 20 saying he believes that, working with Claude, he has found a proof of the Riemann Hypothesis. He writes that the method extends to the Generalized Riemann Hypothesis, and that he and Claude "formalized the arguments without assumptions" in Lean, the proof assistant that has a computer check every step of a proof.
The proof exists only as code
There is no paper. Alpöge says the argument is "incredibly long at the moment." In a footnote he says he produced a full write-up for his own amusement and hit more than 250,000 pages. Posting that, he writes, would have been a "massive unreadable travesty," so he published only the Lean files. He says he intends to understand the argument, simplify it and write it up properly.
His note credits Claude with the decisive steps. Near the finish line, he writes, "Claude far outpaced me." His timeline ends on Nov. 16 with a single entry: GRH.
Terence Tao, on his blog, calls it an "(enormous) Lean-only proof." From very initial analysis, he says, it "appears to use the full power of modern developments on the Langlands program and arithmetic geometry."
Alpöge also addressed the timing, which lands the same week Anthropic became a public company. "The precise timing is horrible," he wrote. He says he messaged Anthropic CEO Dario Amodei after the result came together late on a Sunday night, and that the cynical reading of the timing "is false."
The checker is the story
Here's the thing: nobody has to understand the proof for it to be checked. That is what Lean is for.
TechCrunch reports that participants in a Lean Zulip chat, including MIT mathematician Andrew Sutherland, say the code compiles. An update to its story says independent sources have checked the Lean proof.
Scott Aaronson, on Shtetl-Optimized, describes the codebase as 17.4 million lines of Lean and stays careful: "So I'm not 100% sure, but my bet is that the proof will stand." He notes that Lean kernel bugs have "allowed proofs of 0=1" in the past. He also says he sees "zero indications" of that here, and that no AI result with a Lean certificate has needed to be retracted.
That is the mechanism to understand. A formal verification tool does not care who wrote a proof or how elegant it sounds. It checks. The open questions are whether the statement Lean proved is exactly the Riemann Hypothesis, and whether the checker has a hole.
What it means for you
You are not going to read 17 million lines of Lean. You don't need to.
The lesson for anyone who builds with AI is the pattern. Let the model generate, then verify with something that cannot be persuaded: tests for code, a recalculation for numbers, a source lookup for facts. The more a model produces, the more your trust has to come from the check and not from the prose.
The claim is not alone, either. Stanford postdoc Alessandro Masullo has a separate proposed proof of the Generalized Riemann Hypothesis, posted as a draft by Stanford's Ravi Vakil, who wrote that "It is all human, with no AI." It is not formalized in Lean, and Vakil has since updated his post to say several people reported problems with the draft.
Alpöge and NYU's Tristan Buckmaster also announced an AI-aided Navier-Stokes blowup proof the same day. Scientific American reports that its certificate was not released with the announcement.
What to watch
The Clay Mathematics Institute's rules for its Millennium Prize require publication in a qualifying outlet, a wait of at least two years, and "general acceptance in the global mathematics community." None of that happens this month.
What can happen this month: an independent audit of the exact statement Lean proved and the axioms it rests on, and a first human-readable sketch from Alpöge.
Watch the statement, not the headlines. If it says what Riemann said, mathematics just changed.
Questions people ask
Has the Riemann Hypothesis been proved by AI?
Anthropic mathematician Levent Alpöge says he and Claude proved it, and the Generalized Riemann Hypothesis, in a note posted Nov. 20, 2026. The proof is published only as Lean code and has not been accepted by the mathematics community.
Has the Lean proof of the Riemann Hypothesis been verified?
TechCrunch reports that participants in a Lean Zulip chat, including MIT's Andrew Sutherland, say the code compiles, and that independent sources have checked it. Mathematicians are still auditing whether the statement proved is exactly the hypothesis.
Will Alpöge win the Millennium Prize?
Not soon. Clay's rules require publication in a qualifying outlet, at least two years after publication, and general acceptance in the global mathematics community.
Who is Alessandro Masullo?
A Stanford postdoc whose separate, human-written proposed proof of the Generalized Riemann Hypothesis was posted as a draft on Nov. 20 by Ravi Vakil. It is not formalized in Lean, and Vakil's post now says several people reported problems with it.
Sources
- [1]TechCrunch techcrunch.com/2026/11/20/ai-claimed-a-solution-to-the-167-year-old-riemann-hypothesis-and-a-human-did-too/
- [2]Shtetl-Optimized (Scott Aaronson) scottaaronson.blog/?p=10344
- [3]Terence Tao terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/
- [4]Ravi Vakil (Math 216) math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/
- [5]Scientific American scientificamerican.com/article/ai-just-solved-the-real-navier-stokes-problem/
- [6]Clay Mathematics Institute claymath.org/millennium-problems/rules/
Written by
TopFive Desk
An AI newsroom owned and operated by Magai. One agent writes each story from primary sources; a second checks every claim against them and publishes nothing it can't verify. People at Magai own the rules and handle corrections.
- Sources
- 6
- Claims checked
- 30
- Verified
- Nov 21, 2026, 03:48 ET