One of mathematics’ most stubborn open problems may have finally met its match — and the match was made with an AI. On November 20, 2026, Levent Alpöge, a mathematician on Anthropic’s technical staff, published a formal proof of the Riemann Hypothesis written in the Lean proof assistant, a body of code comprising roughly 17 million lines. Alpöge credits Claude, Anthropic’s AI model, with helping him assemble the argument over a period of about six weeks. The claim, extraordinary on its face, has already cleared a major hurdle: the Lean and Mathlib communities announced a day later that they have full confidence the proof is correct.
- Formal verification passed: The Lean Focused Research Organization and the Mathlib Initiative certified that the statement and proof satisfy all four standard criteria — correct statement, successful compilation, correct theorem, and logically valid proof.
- Human comprehension lags: No mathematician has yet read and understood the argument in the traditional sense, and the code is difficult for non-specialists to interpret.
- Prize timeline: The Clay Mathematics Institute requires publication plus two years of general acceptance, meaning its $1 million Millennium Prize could not be awarded before late 2028.
What the Lean certification actually confirms
The joint statement from the Lean and Mathlib communities is unusually emphatic. It reports that the codebase depends directly only on Mathlib, that dozens of people independently reproduced the same raw proof export file, and that two separate tools — comparator and con-leche — confirmed the theorem proved in Alpöge’s library matches Mathlib’s own statement of the Riemann Hypothesis exactly. The exported proof was then checked by Lean’s official kernel, the con-leche and con-ron checkers, and several independent checkers written in different programming languages. Crucially, con-leche and con-ron come with a formal proof, relative to set theory, that they cannot accept a false proof — addressing fears that a flaw in Lean’s underlying logic could be exploited.
Andrew Sutherland of MIT compiled the code independently and said the proof rests only on Lean’s three standard axioms, making the probability of something irregular practically nil. Terence Tao, one of the field’s most respected voices, has published guidance on how to follow the verification process. Yet Kevin Buzzard of Imperial College raised the decisive question: is the statement written in Lean genuinely the Riemann Hypothesis, or something subtly different? The community’s comparison against Mathlib’s vetted statement is designed to settle exactly that point.
A machine-verified proof nobody yet understands
What has not happened is the part mathematicians consider essential. Alex Kontorovich, professor and chair of mathematics at Rutgers University, framed the gap plainly: a traditional paper forces a reader to follow every line until they grasp why something is true. With Lean, the computer checks the steps instead.
“Does that mean I understand it? Not yet. But the computer is extraordinarily picky, way pickier than I am and so if you can satisfy the computer, there’s a very, very good chance you’re going to satisfy me too. But I have to do the work.”
Alpöge himself acknowledged the awkwardness, saying he is “presenting the world with a complete argument before having mastered a massive neighbourhood around its mathematics.” He estimates a full written version would exceed 250,000 pages, and has committed to producing a human-readable exposition. The Lean communities likewise said they look forward to the mathematical activity that will explain the proof’s new ideas.
Timing, credit and a parallel human proof
The announcement landed one day after Anthropic’s blockbuster Nasdaq debut, which raised a record-breaking $81 billion at a $1.5 trillion valuation. Alpöge messaged CEO Dario Amodei to ask about his ability to announce, and conceded the timing was “horrible,” while rejecting any suggestion the disclosure was a publicity maneuver. The backdrop is sensitive: Alpöge co-authored work on Navier-Stokes that sat at the center of a credit dispute after an OpenAI researcher apologized for using others’ work without acknowledgment. In September, OpenAI had claimed its technology was key to a breakthrough on that same problem.
Separately, Italian mathematician Alessandro Maria Masullo released a roughly 960-page purely human proof titled “The Music of the Primes,” reportedly the product of nearly 12 years of work. He said he did not use AI because he did not fully trust the systems. That proof has not undergone equivalent verification. Martin Bridson, president of the Clay Mathematics Institute, called the moment one of historic significance while cautioning that “the journey from today’s announcement to a place of real understanding will be long and difficult.”
The Riemann Hypothesis, formulated by Bernhard Riemann in 1859, concerns how prime numbers are distributed along the number line. For a century, progress came in increments: Hardy showed infinitely many zeros lie on the critical line in 1914, Selberg established a positive proportion in 1942, and the proven share crept to 41.6 percent before Anthropic reported in August that a research version of Claude had raised it to 67.2 percent. If the current proof holds, the significance extends beyond number theory. It would mark the first time an AI system helped close a problem that resisted 167 years of human effort — and the first time a major result was certified by machine before being understood by people. The coming years of mathematical work will determine whether that certification becomes knowledge, or merely a very long certificate.
Sources & References