An artificial intelligence system from Anthropic has transformed Fermat's Last Theorem from a human-checked masterpiece into a proof a computer can verify end to end in just 11 days. A task once expected to take years was completed with the help of dozens of Claude agents working in the Lean formal language.
The theorem itself was first proven by Andrew Wiles in the 1990s, building on work with Richard Taylor. What Claude did was different: it did not discover a new solution, but converted the accepted human proof into a fully formal version that a machine could inspect step by step without relying on expert judgment.
According to Anthropic, the process produced around 13 million lines of Lean code and more than 30,000 intermediate theorems, with nearly 29,500 used in the final structure. The scale of the result highlights how AI is moving beyond problem-solving and into rigorous mathematical verification.
This matters because formal proof checking could become a powerful tool for mathematics, where long arguments often depend on shared assumptions and hidden shortcuts. A computer-based verifier can flag every missing step, making complex results easier to test, compare, and reuse.
Researchers such as Kevin Buzzard at Imperial College London have been working on similar formalization efforts, but the speed of Anthropic's system suggests a new phase for mathematical research. The challenge now is not only correctness, but also building reusable libraries that future proofs can grow from efficiently.
As AI systems become better at formal reasoning, mathematics may gain a tireless digital reviewer capable of supporting discovery at scale. That could reshape how future theorems are written, checked, and shared across science.
