Anthropic mathematician claims to prove Riemann hypothesis, the 167-year-old puzzle behind the pattern of prime numbers, and the math world is now checking the work. Levent Alpöge announced on Friday, Nov. 20, 2026, that he believes he and Claude, Anthropic’s AI system, found a proof. No outside expert has confirmed it yet, so the claim remains unverified.
Here is what is established, what is still open, and what could happen next.
What Alpöge Claimed on November 20
Alpöge, a member of Anthropic’s technical staff, posted a four-page note on his personal website. His central sentence is carefully hedged: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.”
He says the same method also settles the Generalized Riemann Hypothesis, a broader version that covers a larger family of mathematical functions. He expects it to reach an even more general version with more time.
Alpöge said the breakthrough came late on Sunday, Nov. 15. He then asked Anthropic CEO Dario Amodei about his ability to announce it, as the company prepared for its stock market debut. Anthropic raised a record $81 billion at a $1.5 trillion valuation that week.
The Associated Press reported that Anthropic did not respond to a request for comment on the claim. The AP story, carried by the Boston Globe, is the most detailed mainstream account so far.
What the Riemann Hypothesis Actually Says
Bernhard Riemann proposed the hypothesis in 1859. It concerns a function that encodes information about prime numbers, the whole numbers divisible only by 1 and themselves.
The function has “nontrivial zeros,” and the hypothesis says every one of them sits on a single vertical line in the complex plane. Computers have checked the first 10 trillion zeros, and none has broken the pattern. A proof for all of them has never existed.
The problem is one of seven Millennium Prize Problems that the Clay Mathematics Institute named in 2000, each carrying a $1 million reward. Only one of the seven, the Poincaré conjecture, has been resolved.
How the Proof Was Released
There is no conventional paper. Alpöge released the argument as roughly 17 million lines of Lean code, a programming language that lets a computer check every logical step of a proof. The files came in 14 compressed archive parts.
Alpöge said his work with Claude produced more than 250,000 pages of text. He did not publish them. He offered the short note and the Lean files instead.
He also acknowledged that he is presenting a complete argument before fully mastering the mathematics around it. That candor matters. Even if the code is flawless, humans must still work out why the result is true.
What Is Confirmed and What Is Not
Some early checks have gone well. MIT mathematician Andrew Sutherland said he compiled the Lean code on his own, and the code reportedly builds and relies only on Lean’s standard foundational axioms. Mathematician Elliot Glazer said a preliminary inspection found that the formal statement in the files is the Generalized Riemann Hypothesis.
Much remains open:
- No full independent verification has been announced.
- No human-readable proof exists that experts can study line by line.
- No journal has published or reviewed the work.
- The Clay Institute’s website still listed the problem as unsolved hours after the announcement.
A Lean check shows that a proof follows its own rules. People still have to confirm that the theorem it proves matches the real hypothesis.
How Mathematicians Are Reacting
Reaction has mixed awe with caution. Alex Kontorovich, a Rutgers mathematics professor, said a computer is far pickier than any person, so passing its check makes a good outcome likely. He added that he does not yet understand the proof and still has to do the work.
Terence Tao wrote about the release on his blog and pointed readers to a Lean community discussion thread as the best place to follow the verification effort. Some mathematicians have urged patience until the formal statement and the argument are fully vetted.
Martin Bridson, president of the Clay Mathematics Institute, called the announcement historically significant. He also warned that the road from the announcement to real understanding will be long and difficult.
Anthropic’s Earlier Riemann Hypothesis Work
This is not Anthropic’s first brush with the problem. On Aug. 10, 2026, the company published a research post describing what happened when an unreleased version of Claude was told to take a real stab at the hypothesis.
Claude did not solve it. It did improve a long-standing bound on the share of zeros known to lie on the critical line, raising it from 41.6% to about 67.2%. Alpöge was one of two Anthropic mathematicians who examined that work, and the company said then that it did not expect the techniques to lead to a full proof.
The AP also noted a similar claim in September, when OpenAI said its technology played a key role in a breakthrough on the Navier-Stokes equations. That is another Millennium Prize Problem.
Will It Win the $1 Million Clay Prize?
Not anytime soon. Clay’s published rules call for formal publication, followed by a waiting period of general acceptance in the mathematical community, which the rules set at two years. Verification, publication and expert consensus all have to come first.
For now, the announcement is a claim, not a certified solution.
A Long History of Failed Claims
Skepticism has a long pedigree. Louis de Branges posted a 124-page proof in 2004, and the community never accepted it. Sir Michael Atiyah made a high-profile claim in 2018 that stayed under heavy doubt. In 1945, a proof of the hypothesis’s falsehood appeared in a major magazine, and experts later found it flawed.
The key difference now is the machine-checkable format. A Lean proof leaves little room for hidden gaps in logic. The remaining risk lies in whether the code says what it is supposed to say.
What It Could Mean for Encryption and Daily Life
Readers should not expect any immediate change. A proof would turn many results that mathematicians currently treat as “true if the hypothesis holds” into settled theorems.
It would not, on its own, make factoring large numbers easier, so it does not threaten the encryption that protects online banking and messaging. The broader versions could make a well-known primality test provably fast, which is a technical win for number theory rather than a consumer event.
What to Watch Next
- Lean verification: Experts are testing the code and its final statement in public.
- A readable argument: Mathematicians want a human-understandable explanation of why the proof works.
- Anthropic’s response: The company has not yet commented publicly on the claim.
- Clay Institute status: The hypothesis stays listed as unsolved until the formal process runs its course.
- Journal submission: Peer review would start the long road toward any prize decision.
No timeline exists for any of these steps. Early expert checks could land within days or weeks, while formal recognition would take far longer.
Do you think AI can really crack math’s biggest mysteries? Share your thoughts in the comments and check back for the latest updates.