A mathematician employed by Anthropic claims he has resolved the most famous open problem in mathematics — with Claude as his co-pilot. In a short note posted to his personal site this week, Levent Alpöge states that he believes he has "discovered a proof of the Riemann Hypothesis" while working with the AI model, and that the same method extends to the Generalized Riemann Hypothesis, with the Extended Riemann Hypothesis expected to follow "with more time." The claim, posted as a PDF and discussed across more than a thousand comments on Hacker News, arrives during one of the most consequential weeks in Anthropic's history and marks a stunning escalation in the contest between frontier labs to demonstrate AI-driven scientific discovery. For continuing coverage of breaking AI news, AI Buzz Wire is tracking the verification effort as it develops.

What the Note Claims — and What Was Actually Released

The Riemann Hypothesis, proposed in 1859, concerns the distribution of prime numbers and carries a one-million-dollar bounty from the Clay Mathematics Institute. A proof would consolidate countless results in number theory that mathematicians have long established only conditionally, "if the Riemann hypothesis is true."

Alpöge's note is unusually direct about its own epistemic state. He writes that the argument is "incredibly long at the moment" and that, rather than post what he calls a "massive unreadable travesty" of slop, he decided to "only present Lean" — a reference to the Lean proof assistant. The formalization, which he says is finished, is distributed as a fourteen-part archive linked directly in the note's first footnote. That choice matters: a completed Lean formalization is mechanically checkable by computer, which converts the question from "do experts believe this hand-waving?" into "does the proof assistant accept the term?" — a far narrower target for verification.

The mathematical heart of the argument, as sketched in the note, descends from Stepanov's classical method for resolving the Riemann hypothesis over function fields. The difficulty, Alpöge explains, is that multiplicity — central to Stepanov's approach — is hard to reproduce over the integers, "corresponding to something like differentiating whole numbers." His workaround draws on Fourier-analytic ideas and chains of theta lifts; in one detail he highlights, Claude suggested a chain length proportional to X for the von Mangoldt sum of length X, a choice he describes as the kind that would make "fellow mathematicians fall out of their chairs."

"Claude Flew, and Danced, and Sparkled"

The note is also remarkable for how it describes the collaboration. Alpöge writes that he had planned a decades-long assault on the problem since his student days, deliberately mastering every tool that might prove relevant. In recent months he and Claude worked across a broad mathematical range, but at the decisive moment, he says, the model left him behind: "this time Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line."

He describes guiding the work to a situation analogous to a toy setting that combined two ideas — one already, in his words, "new and shocking" — after which "Claude began to leave me in awe." The tone has drawn both admiration and skepticism online, with some commenters cringing at the anthropomorphization and others noting that, given Alpöge's standing, the claim cannot be casually dismissed.

A Rivalry Between Labs, Playfully Acknowledged

The note sits within a rapidly heating race between AI labs to claim machine-assisted solutions to millennium-class mathematics. In October, Quanta Magazine reported that number theorists had verified an AI-generated proof of the quasi-Riemann hypothesis — a weaker version of the problem — produced by OpenAI, calling it the most significant AI-for-math achievement so far. OpenAI's separate claim on Navier–Stokes became its own controversy, arriving in what participants described as a rushed publication after rumors circulated that an Anthropic-affiliated mathematician had promising results on a Millennium Prize Problem. Alpöge was reportedly one of the two mathematicians working on Navier–Stokes at the time.

His new note extends an olive branch of sorts: he praises "the good taste of the people at OpenAI involved with various mathematical results," citing their earlier work on the Ehrhart conjecture and their handling of the Monge–Ampère equation, and says he "immediately understood Navier–Stokes was going to fall." He also thanks an unusual list of mathematicians and engineers — including Manjul Bhargava, Peter Sarnak, Maksym Radziwiłł, and Jascha Sohl-Dickstein — alongside Anthropic colleagues.

The Timing Problem

Alpöge is candid about the optics. The breakthrough, he writes, came together "late Sunday night," after which he messaged Anthropic CEO Dario Amodei — "Hypothetically, if I were to have . . ." — to work out how he could announce. Anthropic went public with its record Nasdaq listing days earlier, and the note's arrival in IPO week has fueled speculation that the announcement was timed to capitalize on investor attention.

Alpöge pre-empts the accusation head-on: "There is an obvious cynical view, to do with the timing, but it is false." He writes that he "just could not live" with waiting longer, and that Anthropic's leadership and the mathematicians he consulted shared his view on how to proceed. Whether readers accept that explanation or not, the note's own text confirms the announcement was coordinated with the company's leadership.

Verification: Months of Work, Possibly Minutes of Computing

The decisive question is now unusually concrete. Because the formalization is complete and distributed, the mathematical community's first task is narrow: verify the statement being formalized actually is the Riemann Hypothesis, then let Lean's kernel check the proof. Independent experts — including the number theorists who validated the quasi-Riemann result last month — are well positioned to do this quickly. If the kernel accepts, the remaining debate shifts from correctness to understanding: what the argument means, why it works, and how far the method extends, which Alpöge himself calls the real work ahead as he attempts to "impart understanding" of the argument and its "certain simplifications."

If it holds, the implications extend far beyond number theory. A verified proof of a millennium problem co-credited to an AI model would be the strongest demonstration yet that frontier models can produce genuinely new mathematics at the highest level — following a year in which Claude-derived results improved bounds on the zeta function, AI systems generated hundreds of novel mathematical results, and labs began openly racing each other to machine-assisted discoveries. If it fails, the episode still documents something new: a mathematician of serious standing betting his reputation on a human-AI collaboration, and releasing the machinery for the world to check.

Alpöge ends his note on the theme that has defined AI-assisted science this year: "I hope in having the experience of seeing all of number theory's beauty brought together I am just one of the first." Verification, for once, will be fast. Understanding will take longer.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →