November 21, 2026, (Inside AI) — A researcher at Anthropic claims that the company's Claude AI model helped him produce a proof of the Riemann Hypothesis, a 167-year-old problem that carries a $1 million prize and has resisted every human attempt at a solution since 1859.
Levent Alpöge, a mathematician employed by Anthropic, posted the claimed proof on Friday, November 20, written in Lean, a programming language that checks mathematical arguments for logical errors. The proof has not been independently validated, and Alpöge himself acknowledged that the full argument is extremely long and that he does not yet understand every part of it.
The claim lands at the center of a widening debate about what counts as a mathematical proof when machines do the heavy lifting. It also arrives after a year in which AI systems have repeatedly produced results that human mathematicians could not, raising urgent questions about verification, credit, and trust.
Why a Million-Dollar Conjecture Still Matters
The Riemann Hypothesis concerns prime numbers, the whole numbers greater than one that can only be divided evenly by one and themselves. Primes appear at first to scatter randomly along the number line, yet Bernhard Riemann proposed in 1859 that their distribution follows a hidden order tied to a mathematical function called the zeta function.
Specifically, Riemann predicted that all the important zeros of that function, known as non-trivial zeros, lie on a single vertical line in the complex plane, at the position 1/2. That line is called the critical line. If the hypothesis is true, mathematicians gain a far deeper understanding of how primes behave. The Clay Mathematics Institute lists it among its seven Millennium Prize Problems, each worth $1 million for a solution that meets the institute's criteria.
Alpöge described the problem as something that had guided his mathematical ambitions since childhood, saying he had expected to spend decades working toward a solution. He said he worked with Claude over several months and that the model eventually moved far beyond his own pace of reasoning.
"Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line," Alpöge wrote.
He also claimed that the approach could extend to two related problems, the Generalised Riemann Hypothesis and the Extended Riemann Hypothesis.
The announcement follows a pattern of accelerating AI claims in mathematics. In August, Anthropic said an unreleased version of Claude had improved a longstanding bound on the proportion of zeros lying on the critical line, without proving the hypothesis itself. In September, OpenAI said an internal system had produced a proposed solution to the Navier-Stokes problem, another Millennium Prize problem. The following month, OpenAI released 372 families of mathematical results covering hundreds of research manuscripts, while acknowledging they sat at different stages of verification.
Independent verification remains the bottleneck. Lean can check whether a proof is coded correctly, but experts must still confirm that the formalization captures the intended theorem and that no gaps hide in the translation from human reasoning to machine code. Alpöge's admission that he has not fully absorbed the argument complicates that task, because a proof nobody understands is difficult to defend even if a computer accepts it.
The broader pattern suggests AI is shifting from solving textbook problems to generating research-level mathematics. Earlier this year, an OpenAI model disproved a longstanding hypothesis in discrete geometry by producing a counterexample. Anthropic's Claude produced the first complete computer-verified formalization of Fermat's Last Theorem, a result already proved by Andrew Wiles in 1995. Microsoft researchers used human-directed AI to help resolve the 31-year-old KLS conjecture, which has implications for statistics, optimization, and computing.
Each of these results required human mathematicians to frame the problem, judge the output, and confirm the reasoning. The Riemann claim tests whether that division of labor still holds when the machine supplies not just a step but the entire architecture of an argument.
Alpöge framed the result in expansive terms. "One dawn across science of something far greater," he wrote, describing the discovery as a possible pointer toward a wider shift in scientific research.
Whether that dawn arrives depends on what happens next. The Clay Mathematics Institute has not commented on the claim, and no independent team has yet reported a verification. Mathematicians will now examine the Lean code, check the formalization against the original conjecture, and decide whether a machine-assisted proof that its own author only partly understands can satisfy the standards the field has applied for centuries.
Until that scrutiny concludes, the Riemann Hypothesis remains open, and the most consequential mathematics story of the year remains a claim rather than a theorem.
Sofia spent years studying how machines parse human language before pivoting to cover the systems now rewriting it. She reads technical papers the way others read news and looking for what the numbers are actually saying beneath the headlines.