What happened?

Levent Alpöge, a mathematician working at Anthropic, has claimed to have proven the Riemann hypothesis with Claude's support, one of the most deeply entrenched unsolved problems in mathematics. First posed by Bernhard Riemann in 1859, the hypothesis concerns the zeros of the zeta function, which governs the distribution of prime numbers. It is one of the Clay Mathematics Institute's seven Millennium Prize Problems, carrying a one-million-dollar reward if verified. So far, the claim has not undergone formal review by independent mathematicians.

Why it matters

AI playing a role in generating mathematical "solutions" isn't new. DeepMind's AlphaProof came close to gold-medal performance on International Mathematical Olympiad problems in 2024, and figures like Terence Tao have spent months testing AI-assisted proofs using Lean-like formal verification tools. But in none of those cases was a model said to have solved a century-old problem on its own — they all describe a process framed and verified step by step by a human mathematician. Alpöge's claim appears to follow the same pattern: the original mathematical intuition and strategy came from the human, while Claude took on a supporting role — handling computation, step-checking, and rapidly testing possible inferences. That distinction matters, because the headline "Claude solved the Riemann hypothesis" says something very different from "a mathematician produced a proof draft using Claude as an assistant."

The real issue here isn't the claim itself, but whether it can be verified. Formal acceptance in mathematics comes only after a proof is published and subjected to a long period of line-by-line scrutiny by independent reviewers. History counsels caution here: Shinichi Mochizuki's 2012 announced proof of the ABC conjecture has remained disputed for over a decade, with figures like Peter Scholze and Jakob Stix arguing there's a fundamental gap in the proof, which Mochizuki himself does not accept. Even Andrew Wiles's 1993 announcement of a proof of Fermat's Last Theorem had a gap found by reviewers; the fix wasn't completed until a year later, with the help of another mathematician. That risk is even greater in long proofs with AI contributions: large language models can produce small but proof-invalidating errors across chains of reasoning hundreds of steps long, and because the text reads fluently and persuasively, those errors can slip past human reviewers too.

Who does this affect, and how?

For Anthropic, the claim carries significant PR value, strengthening its position in the ongoing "AI is solving math problems" narrative alongside OpenAI and Google DeepMind. But it also carries reputational risk: if the proof collapses, the company's name will be attached to headlines about AI failing at mathematics. For the mathematics community, the outcome is more neutral: attempts like this will keep increasing, but peer review standards won't change. The only Millennium Prize Problem formally solved to date is Grigori Perelman's proof of the Poincaré conjecture — and even that proof, published in the early 2000s, went through years of verification before the Clay Institute awarded the prize in 2010.

That points to a broader pattern: in mathematics, the gap between "solved" and "accepted" grows in proportion to how important the problem is. For a problem as central to prime number theory as the Riemann hypothesis, that gap could stretch even longer than it did for the ABC conjecture — because whether the hypothesis turns out to be true or false directly affects hundreds of other results in number theory, and reviewers will scrutinize every line all the more rigorously because of it.

What's next?

Three things are worth watching: whether the full text of the proof gets published and reaches independent mathematicians, whether figures like Terence Tao weigh in on it, and whether the proof gets tested in formal verification systems like Lean. The Clay Institute's official process could also take years — the institute requires that a proof be published in a respected journal and remain generally accepted for a period of time before any prize is awarded.

Before we can say Claude genuinely produced an original piece of mathematical insight here, we need to see whether this claim survives peer review. The ABC conjecture case shows that an announcement of being "solved" isn't the same thing as scientific acceptance — a distinction that shouldn't be lost on a problem as legendary as the Riemann hypothesis.