Introduction

A mathematician employed by Anthropic has announced a potential solution to the Riemann hypothesis, one of the most famous unsolved problems in mathematics. At nearly the same time, a separate researcher at Stanford put forward a human-crafted proof of a related conjecture. The twin announcements have reignited debate about the role of artificial intelligence in advanced mathematical research.

What Happened

Levent Alpöge, a mathematician at the AI company Anthropic, said he used the Claude language model to help develop a proof of the Riemann hypothesis, a problem first posed in 1859 and recognized as one of the Millennium Prize challenges. The conjecture concerns the distribution of prime numbers and suggests an underlying pattern in their appearance. Alpöge’s result is written in Lean, an open-source programming language that allows mathematical proofs to be checked by computers. Independent reviewers, including mathematicians on the Zulip discussion platform and MIT researcher Andrew Sutherland, have confirmed that the Lean code compiles successfully.

Alpöge described the process as a collaborative effort with Claude, saying the model helped carry the argument to its conclusion. He also notified Anthropic CEO Dario Amodei about the timing of the announcement, which coincided with Anthropic’s debut as a publicly traded frontier AI laboratory.

Separately, Stanford postdoctoral researcher Alessandro Masullo released a proof of the General Riemann hypothesis. Ravi Vakil, a Stanford professor and president of the American Mathematics Association, shared the work and emphasized that it was produced entirely without AI. Masullo said the project represented roughly 12 years of effort. Because Masullo’s proof has not yet been formalized in Lean, the mathematical community has not been able to verify it computationally.

Alpöge and NYU mathematician Tristan Buckmaster also published a separate proof concerning the Navier-Stokes equations, another major problem in mathematical physics.

Why This Matters

The Riemann hypothesis has resisted proof for more than 160 years, and a confirmed solution would have profound implications for number theory and cryptography. If Alpöge’s Lean-verified result holds up under full peer review, it would represent one of the first major mathematical breakthroughs produced with substantial assistance from a large language model.

At the same time, the emergence of Masullo’s human-only proof highlights a parallel narrative. Some researchers see it as a reminder that traditional mathematical insight remains vital, even as AI tools grow more capable. The fact that both announcements arrived within days of each other underscores how quickly the frontier of computational mathematics is moving.

Key Takeaways

  • Levent Alpöge of Anthropic claims an AI-assisted solution to the Riemann hypothesis, formalized in the Lean proof assistant.
  • Independent mathematicians have verified that Alpöge’s Lean code compiles, though full human understanding of the proof is still developing.
  • Stanford postdoc Alessandro Masullo released a separate, human-only proof of the General Riemann hypothesis after about 12 years of work.
  • Masullo’s proof has not been checked in Lean, so its validity remains unconfirmed by computational verification.
  • Alpöge and collaborator Tristan Buckmaster also shared a new result related to the Navier-Stokes equations.
  • The announcements arrive as Anthropic completes its transition to a publicly traded AI lab.

Conclusion

The mathematics community now faces the task of examining two very different paths to the same historic problem. Alpöge’s machine-readable proof offers immediate computational verification, but researchers still need to translate it into ideas that humans can teach and build upon. Masullo’s traditional proof offers a human narrative, but it awaits formal validation. Together, the two efforts illustrate a turning point in which artificial intelligence and human ingenuity are advancing side by side—even if neither result has yet been universally accepted as the final answer.