An Anthropic mathematician says he and the company’s Claude AI have done what generations of the world’s best minds could not: solve the Riemann Hypothesis.
Levent Alpöge published the extraordinary claim Friday, declaring that the 167-year-old problem has been proved and that the argument has been formalized in Lean, a computer proof assistant.
If the claim survives independent scrutiny, it would be one of the most important mathematical breakthroughs in modern history and could open a path toward the $1 million prize attached to the problem.
But that final word matters: this is still a claim. The mathematical community has not yet accepted Alpöge’s proof, and the Clay Mathematics Institute continues to list the Riemann Hypothesis as unsolved.
In Alpöge’s four-page note, he opens with a carefully worded but unmistakable declaration: he believes that working with Claude produced a proof of the Riemann Hypothesis.
He goes further. Alpöge says the same method extends to the Generalized Riemann Hypothesis and that he expects it may eventually reach the Extended Riemann Hypothesis as well.
His note also supplies a rough research timeline and links 14 separate archive files containing the formal Lean work. Alpöge says the decisive phase came late on November 15, when repeated theta lifts appeared to provide the multiplicity the argument needed, followed by the claimed Generalized Riemann Hypothesis result on November 16.
The original Riemann Hypothesis was proposed by German mathematician Bernhard Riemann in 1859. It concerns the hidden structure behind the distribution of prime numbers, the indivisible building blocks that sit at the foundation of number theory.
The Clay Mathematics Institute explains that the hypothesis predicts where the non-obvious zeros of the Riemann zeta function must lie: on a vertical line whose real part is one-half.
That statement sounds compact. Proving it for every relevant zero has resisted mathematicians for more than a century and a half.
Computers have checked the first 10 trillion relevant solutions, according to Clay. Every one tested fits Riemann’s prediction, but no finite mountain of successful checks can replace a proof that covers all cases forever.
Clay named the hypothesis as one of seven Millennium Prize Problems in 2000, each carrying a $1 million award for a qualifying solution. Only the Poincaré Conjecture has been removed from the institute’s unsolved list, which shows just how rarely these famous problems yield.
That is the wall Alpöge says he and Claude have finally broken through.
Alpöge’s account of the collaboration is striking because he does not describe Claude as a glorified calculator or a passive assistant.
He says he guided the project toward the right mathematical setting, but that Claude then surged ahead near the finish line. His note describes the model as supplying technical ideas while the two combined different approaches that had first appeared in a simplified toy setting.
One of the central ideas, as Alpöge tells it, involved chaining mathematical objects called theta lifts. For a von Mangoldt sum of length X, he says Claude proposed using a chain of length equal to the square root of X.
To professional mathematicians, the significance will lie in whether the formal definitions, assumptions and deductions really match the classical hypothesis. To everyone else, the larger story is already plain: a human mathematician says an AI system supplied crucial technical ideas during an attack on perhaps the most famous open problem in pure mathematics.
Alpöge says the collaboration unfolded over roughly six weeks. The timeline in his note begins on October 6 and reaches a claimed Generalized Riemann Hypothesis result on November 16.
He also acknowledges that the argument is enormous and not yet presented as an ordinary paper that mathematicians can comfortably read from start to finish.
Instead, Alpöge released a massive set of Lean files and says the formalization has been completed without assumptions. He jokes that the human-readable version would run beyond 250,000 pages, calling it a “book” only after Claude had used the term.
The Lean organization describes Lean as both a programming language and a proof assistant. Mathematicians can state definitions and theorems precisely, then ask Lean’s small trusted kernel to check whether every logical step follows.
That gives Alpöge’s claim more substance than an AI simply producing a persuasive-looking answer in a chat window. A genuine Lean proof that compiles against the intended definitions would provide machine-checked evidence that its formal steps are logically valid.
It would not, by itself, end every question.
Independent experts still need to verify that the code builds as claimed, that the formal theorem corresponds exactly to the classical Riemann Hypothesis, that no unrecognized assumptions or flawed definitions were smuggled into the statement, and that the mathematical architecture is understood well enough to earn broad acceptance.
That is why Alpöge’s own wording begins with “I believe” instead of announcing that the prize has already been awarded.
The broader Generalized Riemann Hypothesis claim is also part of Alpöge’s account, and he says the work has been formalized in Lean.
The $1 million is not immediate, either. Under the Clay Mathematics Institute’s prize rules, a proposed solution must be published in a qualifying outlet, at least two years must pass, and the work must receive general acceptance throughout the global mathematics community before Clay will consider it.
So even a correct proof enters a long public verification process. A Friday announcement cannot skip that process, no matter how remarkable the code or how powerful the AI behind it may be.
Clay does not accept proposed solutions through direct submission. Its rules are designed to let the mathematical world test, challenge and absorb a result in public before the institute decides whether the prize conditions have been met, placing community acceptance alongside formal publication and elapsed time.
Alpöge’s background makes the claim impossible to dismiss as random internet boasting. His personal academic site identifies him with Anthropic and Harvard’s Society of Fellows, and his work centers on advanced number theory.
He writes that the Riemann Hypothesis has been a guiding problem since his student days and that he spent years gathering tools he thought might someday matter to it.
That human preparation is part of the story. Claude did not wake up in a vacuum and announce a theorem; Alpöge describes years of mathematical judgment meeting a model capable of exploring and executing technical ideas at extraordinary speed.
If the proof fails, the episode will still become a valuable test of where formal AI-assisted mathematics can go wrong. If it holds, mathematics and artificial intelligence may have crossed a line the world will remember.
A problem posed in 1859 may have met its answer in 2026 through a partnership between a mathematician and a machine.
Now comes the hardest part: proving to everyone else that the proof is truly what it claims to be.
Continue reading...
[ H/T WLT Report ]