
Riemann Hypothesis Proof Claimed With Claude After 167 Years
Levent Alpöge, a mathematician at the artificial intelligence company Anthropic, claimed on Friday, 20 November 2026, that he has found a Riemann hypothesis proof while working with Anthropic’s Claude model. The claim appeared in a short statement on his personal website, and the full argument was posted only as computer code in the Lean programming language.
No paper has been published, and no independent researcher has confirmed the result yet. The claim is therefore a claimed proof, not a settled one, and the checks that would settle it have not happened.
Riemann Hypothesis Proof Claim — Quick Reference Summary
| Detail | What is known |
|---|---|
| Claimant | Levent Alpöge, mathematician employed by Anthropic |
| Claim date | Friday, 20 November 2026 |
| Problem | The Riemann Hypothesis, proposed by Bernhard Riemann in 1859 |
| Age of the problem | 167 years |
| Tool credited | Claude, Anthropic’s large language model |
| Where it appeared | A four-page unsigned note titled “Remark” on alpo.ge, Alpöge’s personal website |
| Form of the proof | Lean code only, split into 14 compressed archive parts; no readable paper |
| Extra claim | The same method is said to prove the Generalized Riemann Hypothesis |
| Verification status | Not publicly verified — nobody outside the author’s circle has confirmed the code builds or states the hypothesis correctly |
| Prize attached | $1 million from the Clay Mathematics Institute as one of the seven Millennium Prize Problems |
The 167-Year-Old Problem Behind the Claim
Bernhard Riemann proposed the hypothesis in 1859. It concerns the zeros of the Riemann zeta function, a function that encodes how prime numbers are spread among the integers.
The hypothesis says every nontrivial zero lies on one vertical line in the complex plane, where the real part equals one half. If that holds, the primes are distributed about as regularly as they possibly could be, and thousands of results in number theory that currently assume the hypothesis would stand on firm ground.
Progress was so rare that most leading experts spent little time attacking the problem directly, because no clear path to a proof existed. The first real crack appeared in October 2026, when one result in a large batch of machine-produced mathematics from OpenAI made significant progress on the distribution of the primes.
What Alpöge Posted — and What He Did Not
The statement on Alpöge’s website opens with a direct line: “I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis.” Scientific American’s report on the claim describes the same statement and notes that the proof was posted coded in Lean, a language built so a computer can check every logical step.
The release is unusual. There is no journal paper and no readable write-up; the author says the argument is too long and unfinished to present in readable form, and a human-readable version is promised later.
The note itself carries no author’s name, and Alpöge has not announced it on social media. It includes a 20-step timeline, said to have been compiled by Claude, running from early October to mid-November 2026, with the Generalized Riemann Hypothesis step dated Monday, 16 November 2026.
How the Claimed Argument Is Said to Work
The strategy borrows from function fields, a setting built from polynomials over finite number systems where the analogue of the Riemann Hypothesis has already been proved. One route there is Stepanov’s method, which forces a carefully built auxiliary function to vanish to a very high order, a property called multiplicity.
That method relies on derivatives, which have no natural counterpart for ordinary integers, and that missing piece is one reason the approach never worked for the classical hypothesis. The note’s central idea is a workaround for that gap, and it credits the key step to Claude.
Instead of derivatives, the argument uses a long chain of “theta lifts” that move objects called automorphic forms between symmetry groups, with each lift adding one unit of multiplicity. The remaining steps draw on relative trace formulas, Iwasawa theory, Bessel periods and a weakened form of the Vandiver conjecture.
Why the Claim Cannot Be Taken as Settled Yet
A complete Lean proof that compiles and relies only on standard axioms would be extremely strong evidence. In practice, nobody outside the author’s circle has yet confirmed that the archive builds, or that its final theorem faithfully states the Riemann Hypothesis.
The note gives no build instructions, no size for the formalization and no list of the axioms it relies on. OfficeChai’s detailed reading of the note also records the author’s own hedges: the timeline is called rough, the argument may contain strange or suboptimal choices, and some steps hold open questions rather than settled claims.
The author further says the timing is terrible and insists the decision to go public now was his own, after discussion with Anthropic’s leadership and a few trusted mathematicians. Being thanked in the note — Princeton mathematicians Manjul Bhargava and Peter Sarnak and analytic number theorist Maksym Radziwiłł are among those named — does not mean any of them has checked the proof.
Two Earlier AI Results That Are Not This Claim
Readers who followed this story earlier in 2026 may mix up three separate results. Anthropic’s earlier Claude effort improved a related lower bound — the guaranteed share of zeta zeros known to sit on the critical line — from about 41% to about 67%, and Anthropic itself did not present that as a proof of the hypothesis.
Six weeks before this claim, OpenAI released a paper titled “The Quasi-Riemann Hypothesis” among 722 machine-produced manuscripts. That result narrowed the region where zeros could hide but stopped far short of the hypothesis itself.
The new note claims the full hypothesis and its generalized form for Dirichlet L-functions, and says the Extended Riemann Hypothesis may follow with more time. That is a much larger claim than either earlier result, which is exactly why outside verification matters before anyone treats the problem as closed. For readers tracking how fast these systems are moving, our earlier look at Microsoft’s Project Helix console plans shows the same pattern of rapid, headline-sized technology claims arriving before independent testing.
The Mathematician’s Track Record and the Same-Day Second Claim
Alpöge is not an unknown name making an extraordinary claim. In July 2026 he used Claude Fable to disprove the Jacobian conjecture, and in October 2026 he shared a Claude-produced disproof of the Exponential Time Hypothesis.
On the same day the Riemann note surfaced, Alpöge and New York University mathematician Tristan Buckmaster also posted a claimed proof that the Navier-Stokes equations can blow up without any external force. If both held up, that would be two Millennium Prize Problems addressed in a single day.
Extraordinary claims about the Riemann Hypothesis have a long history of falling apart under scrutiny. A strong track record is a reason to examine this one seriously, not a substitute for examining it.
What Happens Next for the Claimed Proof
The first test for this Riemann hypothesis proof is mechanical. Outside researchers need to download the 14 archive parts, build the Lean project, and confirm that the final theorem is a faithful statement of the Riemann Hypothesis resting only on standard axioms.
If the code passes that test, the problem would arguably be settled in a formal sense even before anyone fully understands the argument. If it fails to build, or the theorem says something weaker than the hypothesis, the claim falls at the first hurdle.
There is also a wider debate about proof style. A Lean-only release risks arriving as a sealed black box that tells mathematicians a statement is true without helping them understand why, a concern Terence Tao has raised about machine-produced proofs, and it sits uneasily with calls from leading mathematicians for slower, properly written-up announcements. Until the checks are done, the honest status is an unsigned, unannounced note making the biggest claim in mathematics. Readers can compare the pace of change with our report on Archive of Our Own returning after its 17-day outage, another case where a community waited for verified facts before drawing conclusions.
Questions Readers Are Asking About the Claim
Has the Riemann Hypothesis been proved?
Not publicly verified. A Riemann hypothesis proof has been claimed by Levent Alpöge, working with Claude, and posted as Lean code on 20 November 2026, but no independent researcher has yet confirmed that the code builds or that its final theorem states the hypothesis correctly.
Who is Levent Alpöge?
Alpöge is a mathematician employed by Anthropic, known for earlier machine-assisted results in 2026, including a July disproof of the Jacobian conjecture using Claude Fable. His personal website lists a Princeton PhD thesis and papers on number theory topics such as Hilbert’s tenth problem.
What is Lean, and why does it matter here?
Lean is a programming language and proof assistant in which a computer can check every logical step of a proof. A complete Lean formalization that compiles on standard axioms is strong evidence, but the archive still has to be built and inspected by outsiders before it counts.
What does the Generalized Riemann Hypothesis add?
The generalized form extends the claim from the Riemann zeta function to a whole family of related functions called Dirichlet L-functions. The note says the same method proves that wider statement, with the Extended Riemann Hypothesis expected to follow with more time.
Is there prize money for solving it?
Yes. The Riemann Hypothesis is one of the seven Millennium Prize Problems, carrying a $1 million reward from the Clay Mathematics Institute. A claimed proof would still need to survive the institute’s verification process before any prize was awarded.
Claims Worth Checking Before Sharing
Claim: “Claude solved the Riemann Hypothesis in August.” That earlier Anthropic result raised a proven lower bound from about 41% to about 67% of zeros on the critical line. It was real progress and was never presented by Anthropic as a full proof.
Claim: “The proof has been peer reviewed.” No paper has been published and no peer review has taken place. The argument exists as Lean code in 14 archive parts, and outside verification has not yet been reported.
Claim: “Famous mathematicians have endorsed the proof.” Several leading mathematicians are thanked in the note, but thanks are not verification. None of those named has been reported as having checked the formalization.




