Blog
Anthropic's Claude agents produce the first complete computer-checked proof of Fermat's Last Theorem
Anthropic reports that Claude agents produced the first end-to-end, computer-checked proof of Fermat's Last Theorem in Lean in about 11 days. The formalization was checked with Lean's three standard axioms and compared against Mathlib's statement of the theorem.
A machine-checked FLT, not a new proof idea
On September 4, 2026, Anthropic announced what it describes as the first complete computer-checked proof of Fermat's Last Theorem (FLT). Claude agents wrote the formalization in the Lean proof assistant, working largely autonomously over 11 days. The result is a verification artifact: Andrew Wiles's 1995 proof (with Richard Taylor) remains the classical mathematics; Claude's contribution is an end-to-end Lean encoding that a computer can check.
Anthropic frames the novelty as verification rather than discovery. Unlike recent AI work on the Riemann hypothesis that produced novel mathematics, the company says what is new here is checking a known argument the way a calculator checks a computation. FLT is a sharp test case: Wiles's published proof ran to 129 pages, and human verification historically exposed a critical gap that took roughly a year to repair before the corrected proof appeared in May 1995.
Scale of the Lean artifact
Along the way, Claude wrote about 13 million lines of Lean and proved about 29,500 intermediate theorems used in the final proof (Anthropic also reports about 30,300 theorems proved during the campaign). Dozens of Claude agents collaborated to define concepts, prove lemmas, and assemble harder statements.