Anthropic Says Claude Encoded Fermat’s Last Theorem in 13 Million Lines of Verifiable Code

Anthropic says its Claude AI system spent 11 days converting a proof of Fermat’s Last Theorem into 13 million lines of code that a computer can check independently. The claim concerns one of mathematics’ best-known problems, which remained unresolved for more than three centuries before Andrew Wiles announced a proof in the 1990s. Rather than presenting the result only in conventional mathematical notation, Claude reportedly produced a computer-checkable version designed to remove the need for human trust in the proof’s validity. Decrypt reported the development on September 5, 2026. The announcement highlights a potential use of artificial intelligence in formal mathematics: translating complex arguments into code that verification software can inspect. The supplied information does not indicate that Claude discovered a new proof or that the underlying theorem was previously unsolved; it states that the model turned Fermat’s Last Theorem into a lengthy, independently checkable coded proof.