Anthropic announced that its Claude AI model proved Fermat’s Last Theorem in 11 days. The Lean-formatted formal proof was completed almost entirely autonomously, generating 13 million lines of mathematical code that can be verified line by line by a computer.
The mathematical puzzle was first penned by Pierre de Fermat in the margin of a math book in 1637 and remained without a complete proof for 358 years. Mathematician Andrew Wiles finally proved it in 1995 through a 129-page paper grounded in modern mathematics. Wiles’s proof was initially rejected by reviewers after a logical gap was discovered, before being corrected over nearly a year alongside Richard Taylor. Claude did not merely review those notes, but constructed a formal proof from the most fundamental axiomatic level.
How Dozens of Agents Worked Without Colliding
The project was conceived by Tianyi Peng from a Columbia University team, who built an AI formalization pipeline using a model equivalent to Claude Fable 5.1. Dozens of Claude agents were deployed to work in parallel to solve over 30,000 supporting theorems, consuming billions of tokens throughout the process.
The key to system coordination relied on Prove2Me, a tool that provides each agent with a shared, real-time work queue. This mechanism ensured no agent duplicated proofs or deviated from the core logical chain.
The final output produced code five times larger than Mathlib, the shared mathematical library maintained by global mathematicians, equivalent to roughly 160 novels. The complete proof manuscript has been released on GitHub and is available for public inspection.
Mathematician Validation and Its Impact on Crypto Security
Kevin Buzzard, an Imperial College London mathematician leading a similar formalization project targeting a 2029 completion, has reviewed and endorsed Claude’s work. Buzzard stated the proof is valid with no assumptions beyond fundamental mathematical axioms.
For the crypto and Web3 industry, this AI breakthrough in formal verification has direct implications for cryptographic strength and blockchain security. Mechanical verification systems enable developers to catch logical vulnerabilities in smart contracts and consensus protocols much faster, mitigating exploit risks before code is deployed on mainnets.
With millions of lines of logical reasoning verified in mere days, digital security auditing standards are entering a machine-verified era.
Via Decrypt.
Disclaimer: This article is for informational and educational purposes only, not financial advice. Cryptocurrency assets are highly volatile and carry significant risk. Always do your own research (DYOR) and never invest more than you can afford to lose.
