Anthropic Mathematician Claims Proof of Riemann Hypothesis, Built With Claude
A four-page note and 14 Lean archive files assert both the hypothesis and its generalized form

Levent Alpöge, a mathematician at Anthropic who also holds an affiliation with Harvard University, posted a note to his personal website on November 20, 2026, asserting that he had, working with Claude, discovered a proof of the Riemann Hypothesis. The note extended the claim to the Generalized Riemann Hypothesis as well, covering a broader class of mathematical objects called Dirichlet L-functions. The announcement arrived one day after Anthropic began trading on the Nasdaq Global Select Market under the symbol ANT.
"I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis," Alpöge wrote in the posted note.
The Riemann Hypothesis, proposed by Bernhard Riemann in 1859, holds that all nontrivial zeros of the Riemann zeta function lie on the critical line, the vertical line in the complex plane where the real part equals one-half. Its resolution would settle longstanding questions about the distribution of prime numbers. The Clay Mathematics Institute named it one of seven Millennium Prize Problems in 2000, attaching a $1 million award that remains unclaimed.
Alpöge was candid about the limits of his own understanding of the submitted argument. "I am presenting the world with a complete argument before having mastered a massive neighbourhood around its mathematics," he wrote.
The Proof Strategy: Stepanov's Method Transferred via Theta Lifts
The core of Alpöge's claimed proof attempts to transplant a technique known as Stepanov's method, originally developed in the 1970s to prove the Riemann Hypothesis in the simpler setting of function fields over finite fields, into the harder classical setting involving ordinary integers.
Stepanov's method works by forcing an auxiliary function to vanish to high multiplicity. It relies on derivatives, which have no natural counterpart in the integer setting and have long blocked a direct transfer of the technique. Alpöge's note proposes replacing derivatives with a long chain of theta lifts, a tool from the theory of automorphic forms that moves mathematical structure between different symmetry groups. Each lift in the chain contributes one unit of the multiplicity that the method requires.
The specific length of that chain, approximately equal to the square root of the prime size under analysis, is a parameter that the note credits to Claude. For a von Mangoldt sum of length X, using roughly the square root of X theta lifts appears to make the approach tractable. Additional machinery named in the proof includes relative trace formulas, Iwasawa theory, Bessel periods, and a weakened form of the Vandiver conjecture.
What Lean Compilation Confirms — and What It Does Not
Alpöge released the formalized proof as 14 compressed archive files, together containing approximately 175,000 Lean source files spanning 239 megabytes. The argument runs to four pages in human-readable form. His note states the formalization was produced in Lean 4 using the Mathlib standard library, with no additional axioms beyond those already present in Mathlib.
Lean is an interactive proof assistant that checks the logical validity of each step in a formal argument against its declared axioms. A successful Lean compilation confirms that the proof is internally consistent under those axioms and that the final theorem statement is derived from them without hidden gaps. It does not independently verify that the encoded statement accurately matches the mathematical content the author intended to prove.
Elliot Glazer, a Harvard-trained set theorist and formerly lead mathematician at Epoch AI, where he led the FrontierMath benchmark, confirmed on X that the formalized statement corresponds to the Generalized Riemann Hypothesis, not merely a weaker or technical variant. Glazer also noted that the probability of the result being an artifact of a Lean bug is below one percent. As of Alpöge's announcement, he said, the Lean compilation was still running and had not yet completed independently.
Expert Reactions: Genuine Surprise, Principled Caution
The submission drew immediate attention from mathematicians.
Thomas Bloom, a Royal Society Research Fellow at the University of Manchester, described his reaction as surprise at the scale of the formalization. The proof, if correct, would constitute a human-readable argument of more than 250,000 pages when written out in full, Bloom noted. He added that he was surprised to see the work posted publicly in its current form.
Terence Tao, the Fields Medal recipient whose views carry particular weight in analytic number theory, raised a structural concern about the Lean-only release format. A proof verified exclusively through Lean risks producing a sealed black-box result — one that establishes a theorem as true without giving working mathematicians insight into why it is true. That concern, he said, cuts against calls from a group of Fields medalists who had asked for slower, properly written-up announcements before posting.
The Lean-only format also means that the mathematical community cannot yet assess whether the high-level strategy is genuinely novel or whether it contains a subtle error at the conceptual level that the formal system would not catch.
Historical Precedent and Why This Attempt Differs
Alpöge's earlier work established a pattern of AI-assisted mathematical achievement. On July 20, 2026, the same day as the FIFA World Cup final, Alpöge announced that he and Claude Fable 5 had disproved the Jacobian conjecture, an 87-year-old problem in algebraic geometry first posed in 1939. The result was independently verified by other mathematicians the following day.
Alpöge and Ralph Furman, another Anthropic mathematician, also internally validated an earlier Anthropic result: in August 2026, an unreleased research version of Claude raised the lower bound for the fraction of Riemann zeta function zeros known to lie on the critical line, increasing it from 41.6% to 67.2%. At the time, Anthropic explicitly cautioned that those techniques were not expected to lead to a full proof of the Riemann Hypothesis.
The November 20 announcement represents a qualitatively different claim: not an improvement on a bound but a proposed proof of the full statement.
Cryptographic Stakes and the Clay Prize Timeline
The Riemann Hypothesis underpins the theoretical foundations of several cryptographic protocols. Some encryption schemes depend in part on assumptions about prime distribution that the hypothesis governs, and the Generalized Riemann Hypothesis similarly supports security arguments for certain computational hardness assumptions. A verified proof would not immediately break any deployed system, but it would require a reassessment of the theoretical foundations on which those security proofs rest.
The Clay Mathematics Institute's prize conditions require that any claimed proof be published in a qualifying mathematics journal and then survive a two-year period of review by the broader mathematical community before an award is considered. No prize can be granted until that process runs to completion.
Alpöge closed his note by acknowledging the asymmetry between what has been produced and what can yet be understood: a complete formal argument exists in machine-checkable form, but the mathematical community's ability to verify its meaning in human terms is only beginning.



