Anthropic's Claude AI formalizes Fermat's Last Theorem proof in Lean

How Claude Handled the Logical Complexity of Wiles' Argument

Anthropic announced that its AI model Claude successfully formalized Andrew Wiles' 1995 proof of Fermat's Last Theorem using the Lean theorem prover, completing the task in 11 days. The effort transformed a 350-year-old mathematical argument into over 13 million lines of verified code, marking a significant milestone in AI-assisted mathematics. The project demonstrates how large language models can contribute to rigorous proof verification in pure mathematics.

The formalization process required Claude to translate Wiles' complex, 129-page proof into Lean's precise logical language, a task that involved interpreting subtle mathematical nuances and ensuring every step adhered to the system's strict syntax. Despite the AI's capabilities, the 11-day timeline highlights the immense computational and logical depth of the original proof, which remained unverified in full formal detail until now. Anthropic emphasized that the result is not a new proof but a machine-checkable version of Wiles' landmark achievement.

What Does This Mean for the Future of AI in Mathematics?

Claude had to navigate advanced mathematical concepts including modular forms, elliptic curves, and Galois representations, translating them into Lean's constructive framework. The AI worked under the guidance of human mathematicians who provided strategic direction and validated intermediate steps. Each line of Lean code corresponds to a logical inference, meaning the 13 million lines represent a fully auditable chain of This level of detail surpasses previous formalization efforts in both scale and specificity.

The successful formalization suggests that AI models can assist in verifying highly complex proofs, potentially reducing human error in foundational mathematics. However, the lengthy duration underscores that current AI still requires substantial human oversight for such tasks. Researchers see this as a step toward AI tools that could help explore conjectures or validate proofs in areas like number theory and algebraic geometry. The project also raises questions about how to best integrate AI into the peer review process without replacing expert judgment.

Why did it take 11 days for Claude to formalize the proof? The 11-day duration reflects the immense logical depth of Wiles' proof, which required translating hundreds of intricate mathematical steps into Lean's exacting formal language, despite AI assistance.

Frequently Asked Questions

Is this a new proof of Fermat's Last Theorem? No, this is not a new proof but a formal verification of Andrew Wiles' original 1995 proof, making it machine-checkable for the first time.

What is Lean and why was it used? Lean is a proof assistant and programming language designed for writing mathematically rigorous, verifiable proofs; it was chosen for its strong support in formalizing advanced mathematics.