Skip to content
LIVE
The Executives BriefThe Executives Brief

Anthropic researcher says Claude cracked math's 166-year-old Riemann hypothesis

The claim is unverified and the proof is reportedly so sprawling that checking it could take years, which is now the hard part of AI-driven discovery.

ByLama Al-RashidTechnology Correspondent, The Executives Brief
·4 min read
Anthropic researcher says Claude cracked math's 166-year-old Riemann hypothesis
Executive summary

An Anthropic researcher says the company's Claude model has produced a proof of the Riemann hypothesis, a 166-year-old problem that has resisted mathematicians since 1859. The claim is unverified and the proof is reportedly so sprawling that confirming it could take years, shifting the burden from announcement to verification.

An Anthropic researcher says Claude, the company's AI model, has cracked the Riemann hypothesis, the 166-year-old problem often described as the most famous open question in mathematics. The claim is not a published, peer-reviewed result. It is a researcher's assertion that the model produced a proof, and the proof is reportedly so sprawling that deciphering it could take years. That last detail is the real story. The news is not that a machine may have solved a problem humans could not. The news is that the answer may arrive in a form no human can quickly read, check, or trust.

Start with what is at stake. The Riemann hypothesis, posed by Bernhard Riemann in 1859, is a statement about prime numbers, the indivisible integers that are the atoms of arithmetic. Riemann built a function whose behavior appears to predict how primes are distributed, and the hypothesis says the most important solutions to that function sit on a single straight line. Prove it and you inherit a century of results that mathematicians have established only conditionally, on the assumption that the hypothesis is true. The Clay Mathematics Institute named it one of seven Millennium Prize Problems in 2000, with $1 million attached. Only one of the seven, the Poincare conjecture, has fallen, and the mathematician who solved it, Grigori Perelman, declined the prize.

AI systems have been edging into mathematics for years, and the trajectory has been steep. Google DeepMind's AlphaProof and AlphaGeometry systems reached silver-medal standard at the 2024 International Mathematical Olympiad, solving four of six problems. Those were competition questions with known answers and short solutions. The Riemann hypothesis is a different species of problem: open for 166 years, attacked by the best minds in the field, and resistant to every technique thrown at it. It is also exactly the kind of target where a language model's fluency becomes a liability. These systems generate plausible mathematics at speed, and plausible mathematics is not the same as correct mathematics.

Which is why the phrase "could take years to decipher" matters more than the claim itself. Modern proofs are already enormous. The classification of finite simple groups, completed in the 1980s, runs to tens of thousands of pages across hundreds of papers. A machine-generated proof of a problem this deep could be longer, denser, and written in a style no human would choose. The field has a tool for this: formal verification languages such as Lean, which check every logical step by machine. If the proof can be formalized, verification becomes a matter of compute rather than decades of human reading. If it cannot, the mathematical community is left with a document nobody can audit, which is functionally the same as no proof at all.

For anyone outside the field, the practical stakes are indirect but real. Prime numbers are the raw material of modern encryption, and the distribution of primes is what makes widely used public-key systems hard to break. A proof of the Riemann hypothesis would not crack RSA overnight, but it would sharpen the deepest known facts about how primes behave, and it would validate a large body of work that currently rests on an assumption. The bigger prize is the precedent. If an AI model can produce a genuine proof of a 166-year-old problem, the economics of discovery change: fewer specialists needed per result, faster iteration, and a new kind of scientific output that arrives faster than institutions can review it.

That is why the claim deserves scrutiny rather than either dismissal or applause. Anthropic competes with OpenAI, Google DeepMind, Meta, and others, and a verified mathematical discovery is the strongest possible proof of capability, worth more than any benchmark score. The incentive to announce first and verify later is enormous. The counterweight is that mathematics is unusually unforgiving: a proof is either right or it is not, and the field has spent centuries building institutions, peer review, and now formal verification, to make that judgment independent of whoever is making the claim. Labs that skip that step pay for it in credibility.

Watch three things. First, whether the proof is published in a form mathematicians can examine, and whether it is formalized in a proof assistant. Second, whether the result is independently confirmed, because a claim of this magnitude typically requires more than one group to sign off. Third, whether anyone can explain the idea in human terms, since a proof that no person understands is a strange kind of knowledge: true, perhaps, but not yet usable. The Riemann hypothesis has survived 166 years of the best human effort. If it falls now, the question will not be whether machines can do mathematics. It will be whether we can still understand the answers they hand us.

Executive ActionsLocked

This story's Key Insights and Take-aways are locked.

Create a free account to unlock Executive Actions for one credit.

Register to Unlock

Always free for Executives Club members. Join the Club

More in Technology