Anthropic's Claude Completes First Computer-Checked Proof of Fermat's Last Theorem
The AI model autonomously generated 13 million lines of Lean code to formalize one of mathematics' most famous proofs.
Anthropic has announced the first complete computer-checked proof of Fermat's Last Theorem, marking a significant leap in the application of artificial intelligence to formal mathematics. The proof was produced by the AI model Claude, which worked largely autonomously over an 11-day period.
To achieve the formalization, Claude utilized the Lean programming language, generating 13 million lines of code. During the process, the AI proved 29,500 intermediate theorems to reach the final conclusion. This machine-readable version of the proof allows for verification with absolute mathematical certainty, removing the potential for human error in the review process.
The Legacy of Fermat
Fermat's Last Theorem was first conjectured by Pierre de Fermat around 1637. The theorem states that no three positive integers a, b, and c can satisfy the equation aⁿ + bⁿ = cⁿ for any integer n greater than 2. For over 350 years, the problem remained one of the most elusive and famous unsolved mysteries in the history of mathematics.
It was not until 1995 that Sir Andrew Wiles provided a human-written proof. Wiles' original work spanned 129 pages and required months of rigorous verification by the mathematical community before it was accepted. Formalization is the process of translating such complex human reasoning into a language like Lean, which a computer can verify logically and exhaustively.
Implications for Mathematical Research
This milestone suggests that AI is now capable of formalizing large-scale, complex mathematical proofs that previously required immense human effort. By automating the tedious aspects of proof verification, AI could significantly reduce the burden on human referees who must manually check new submissions for errors.
According to Anthropic, the speed of this achievement demonstrates that it is now possible to formalize large swaths of mathematics. This capability may not only streamline the peer-review process but could also be used to identify existing errors within the broader body of established mathematical proofs, ensuring a higher standard of certainty across the field.
The Future of Formalization
As AI-driven mathematical research accelerates, the industry is moving toward a model where the scale and speed of formalization could lead to the discovery of new theorems. The ability of a model to operate autonomously for nearly two weeks to solve a problem of this magnitude indicates a shift in how complex logic is handled.
Observers will now be watching to see if this approach can be applied to other unsolved conjectures or if the 13-million-line scale of the Lean code presents new challenges for human-AI collaboration in mathematics.