AI Just Claimed to Solve Math's Hardest Problem
Anthropic's mathematician says Claude helped prove the Riemann hypothesis — the most famous unsolved problem in mathematics. If verified, it could permanently change how math is done.
The Proof Is Already Written — Now Comes the Hard Part
Dr. Levent Alpöge did not announce his result on arXiv. He did not send it to colleagues for private circulation before the press conference circuit. He published a computer-checkable proof on his own website, wrapped in Lean — a formal language that strips away ambiguity and forces every step of reasoning into a verifiable chain.
That detail alone signals what kind of moment this is. For centuries, mathematical breakthroughs traveled through informal networks: preprints, seminar slides, coffee room conversations. A proof was considered credible until scrutiny either validated or destroyed it. Alpöge bypassed that entire process by writing a proof a machine can validate directly.
If the verification holds, two things happen simultaneously. The Riemann hypothesis ceases to be open — the $1 million Clay Mathematics Institute prize becomes formally claimable. More quietly, a precedent is set: a proof discovered with substantial assistance from a large language model, structured for automated checking, released for immediate public validation.
Neither outcome is trivial.
Who Alpöge Actually Is
Before dismissing this as an AI hype cycle stunt, the record matters. Alpöge is a dual U.S.-Turkish citizen with an MA from Harvard, a Master of Advanced Studies from Cambridge, and a PhD from Princeton. He works at Anthropic. In July, he posted a counterexample to the Jacobian conjecture — another well-known open problem in algebraic geometry — again crediting Claude’s involvement. The pattern is consistent: he identifies a conjecture, uses an LLM as part of the exploratory process, arrives at a result, and releases a formal proof.
This is not amateur hour.
The Jacobian conjecture work attracted attention precisely because it showed an AI tool helping identify a counterexample in a domain where brute-force computation had never succeeded. The Riemann hypothesis claim, if true, operates at a far higher tier of difficulty. It connects the distribution of prime numbers to the zeros of the Riemann zeta function — a relationship that has resisted proof for 166 years and sits at the core of analytic number theory.
What Claude Actually Did
The source material is deliberately vague about the mechanics. Alpöge wrote: “working with Claude.” That phrase carries a universe of possible meanings. Did Claude suggest a known approach Alpöge then formalized? Did it surface an unexpected connection between distant areas of mathematics? Did it generate an intermediate lemma that required human judgment to integrate?
None of those scenarios diminishes the result. But they do matter for what this case will teach us.
There is a significant difference between an AI that helps you find the right question and an AI that constructs steps no human would have independently considered. The former preserves the traditional understanding of mathematical agency. The latter — if true here — represents a genuine boundary crossing.
The fact that the proof is in Lean complicates both interpretations equally. Lean does not care who contributed which idea. It only checks logical validity. If the formal argument passes, the result stands regardless of how the insight was reached.
Why Verification Is the Actual Story
Two independent groups are reportedly examining the proof. This is the bottleneck now, not the announcement itself. A Lean proof can in principle be checked mechanically, but the ecosystem is still small. Thousands of mathematicians worldwide use Lean daily, but fewer than a handful operate at the frontier level Alpöge appears to have reached. The verification effort may take weeks or months — not because the math is necessarily wrong, but because the community needs to reach consensus on the interpretation.
If the proof holds, the implications extend far beyond prime numbers.
The Three Winners
First: Alpöge and Anthropic. Whether he receives full individual credit or shares it with Claude’s developers, this places both at the center of the most consequential discussion in mathematics in a century.
Second: the Lean community. Formal verification has long been aspirational. A high-profile result derived with AI assistance and checked in Lean provides a template that could accelerate formalization across every branch of pure mathematics.
Third: anyone who has ever watched a mathematician struggle in silence. For generations, the culture around big problems has been solitary brilliance. This moment reframes the narrative entirely.
The Losers Are Cultural, Not Practical
Peer review as an institution loses nothing — verification is still happening, just without gatekeepers. But the cultural authority of the lone genius mathematician does.
There is also a practical concern worth naming. If Claude or its successors become instruments capable of producing or guiding proofs at this level, the cost of entry drops. The barrier shifts from “do you possess deep domain knowledge” toward “do you know how to collaborate with the tool effectively.” That is democratizing in one direction and destabilizing in another. Institutions built on prestige and slow credentialing — universities, journals, prize committees — will need to adapt faster than they typically do.
What Happens Next
Assume the proof verifies. The mathematical consequences are enormous: the Riemann hypothesis sits at the root of modern cryptography, computational complexity, and countless conditional results throughout number theory. Every theorem proven assuming RH remains true. New theorems that depend on the newly established fact will follow within months.
The procedural consequences may matter more. If the next millennium problem is solved the same way — announced on a personal website, encoded in Lean, assisted by an AI system — the entire pipeline of mathematical publication collapses into a single question: can the proof be checked?
Alpöge himself framed this larger than a single result. “What I saw during the course of progress so amazed me,” he wrote, “that I believe it is one dawn across science of something far greater.”
He is not being vague for emphasis. The dawn is real, even if the light has not yet fully broken.