Skip to content
AI EmployeeSuper-AgentsAgent-to-AgentBenchmarksPricingBlog

Anthropic Mathematician Claims Riemann Proof With Claude

At a glanceQuick answers
Who made the claim?
Levent Alpöge, a mathematician who works at Anthropic and is a junior fellow at Harvard’s Society of Fellows, in a note on his own site.
Has the proof been checked?
Not by anyone outside yet. He posted Lean files that a computer can check, but no independent check has been published.
What part did Claude play?
He says he guided the work and Claude supplied key technical steps, and that Claude ‘far outpaced’ him near the end.
Data illustration on a near-white ground: a tall orange stack of paper pages beside a red wave crossing a dashed vertical line again and again, a small teal speech bubble above, and the huge words 250,000+ pages.
Fig 0The note says a generated write-up passed 250,000 pages, so only the Lean files were posted. Made by CellCog's image agent, running GPT Image 2.6.

A mathematician at Anthropic says he and Claude have proved the Riemann Hypothesis. In a four-page note posted on November 20, 2026, Levent Alpöge writes: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” The note went to the top of Hacker News within hours.

It is a claim, not yet a result. Below is what the note says, what has and has not been checked, and why the Lean files are the part that matters.

On this page · 6 sectionsOpen
  1. What the note claims
  2. What has not been checked
  3. Why a Lean proof matters here
  4. The timing
  5. What it means if you hand real work to AI
  6. Sources
Key points5 · 5 min full read
  1. A sheet of paper with a wavy curve crossing a dashed line.
    Levent Alpöge, a mathematician at Anthropic, posted a four-page note on November 20, 2026 saying that he and Claude found a proof of the Riemann Hypothesis, and of the Generalized Riemann Hypothesis too.
  2. Code brackets around a padlock with a check mark.
    He says he and Claude formalized the argument in Lean, a language a computer can check line by line, and he posted only the Lean files, split into 14 parts.
  3. A tall leaning stack of paper pages.
    There is no paper. A footnote says a generated write-up ran past 250,000 pages, and he calls his own short timeline a summary that ‘might make zero sense’ in places.
  4. A magnifying glass over a question mark.
    Nobody outside has reported checking the Lean files yet, and Anthropic has said nothing on X since the note went up. Treat it as a claim until a check is published.
  5. A calendar page beside a clock.
    The note came the day after Anthropic’s stock market debut. Alpöge calls the timing horrible and says the cynical reading of it is false.

§ 01What the note claims

  • The proof. A proof of the Riemann Hypothesis that, he writes, “easily extends much further”, to the Generalized Riemann Hypothesis.
  • The check. “Claude and I have also formalized the arguments without assumptions.” The formal proof is in Lean, linked from the note in 14 parts.
  • No paper. He says producing a readable write-up would take time, so “I decided to only present Lean.” A footnote says a generated write-up passed 250,000 pages.
  • Claude’s part. He says he steered the work toward the right setting and that “Claude far outpaced me near the finish line.”
  • The dates. His own timeline runs from October 6, 2026 to November 16, 2026, the day the Generalized Riemann Hypothesis fell, by his account.
Animation: an isometric platform with a grey Mathematician pillar and a teal Claude pillar; six Lean file blocks slide to a Proof checker box whose light stays grey, an orange ring appears with the label Not yet checked by others, a stack labelled 250,000+ pages stays unopened; then the line Claimed and formalized. Not yet checked.
Fig 1The Lean files go to a proof checker whose light has not turned on yet

§ 02What has not been checked

Part of the claim What the note says What we could check
The proof Found with Claude Only the note itself; no paper
The Lean files Finished, no assumptions Linked in 14 parts; not run by us or reported checked by others
The timeline October 6 to November 16 His own summary, which he says may be rough
Anthropic He consulted leadership No post from Anthropic’s accounts on X
Outside review A few trusted mathematicians consulted No published review
Table 1The claim, and where it stands (read at 15:14 UTC, November 20, 2026)

We did not download or run the Lean files. A check needs a working Lean setup and time, and it is a job for people who can judge whether the formal statement is the real hypothesis.

§ 03Why a Lean proof matters here

Lean is a proof language: a program checks every step, so a proof that compiles does not depend on anyone reading it. That is why a claim with no paper can still be taken seriously. Three things decide it: whether the formal statement really says the Riemann Hypothesis, whether any step leans on an unproved assumption, and whether the files compile. The note says the second holds. The first and third are for others to confirm.

His own words set the bar. He calls parts of his timeline a paraphrase of a summary that “might make zero sense”, and says “the Lean should suffice to clarify everything.”

§ 04The timing

The note appeared the day after Anthropic’s shares began trading on Nasdaq. Alpöge addresses it directly: “The precise timing is horrible.” He writes that he messaged Dario Amodei to ask about announcing, that “there was a constraint to respect”, and that “There is an obvious cynical view, to do with the timing, but it is false”. We have no way to check that, and we report it as his account.

Coverage started fast. One widely shared X post quotes the first press report: “Levent Alpöge, a mathematician and employee at artificial intelligence company Anthropic, claimed on Friday to have proven the most famous problem in all of mathematics using an LLM.”

§ 05What it means if you hand real work to AI

The pattern in the note is the one we see in everyday work: a person who knows where to aim, and an AI that carries the technical load faster than expected. The difference here is the check. A machine-checkable proof is the strongest way there is to show your work, and the question for anyone handing work to AI is the same at any scale: what would let someone else verify it without trusting you?

§ 06Sources

Frequently asked5 questions

Q1What is the Riemann Hypothesis?

A statement from 1859 about where the zeros of the Riemann zeta function lie. If true, it pins down how evenly prime numbers are spread out. It is one of the most famous open problems in mathematics.

Q2Why does a Lean proof matter?

Lean checks every step mechanically. If the files state the real hypothesis, use no unproved assumptions and pass the checker, the proof stands without anyone reading 250,000 pages. Those three things are what checkers will look at first.

Q3Did Anthropic announce this?

No. The note is on Alpöge’s personal site. He says he asked Anthropic’s leadership about announcing it, and Anthropic’s own accounts had posted nothing about it by 15:14 UTC.

Q4Would this win the Clay prize?

The Clay Mathematics Institute offers $1 million for a proof, but only after it is published and broadly accepted by mathematicians, which takes years.

Q5What does this have to do with AI at work?

It is the clearest case yet of a person directing the work while an AI does much of the technical heavy lifting, with a machine check standing in for trust.

Published 20 November 2026 All Category basics →