Anthropic's Claude AI formalizes proof of Fermat's Last Theorem in 11 days

Anthropic AI disclosed that its Claude chatbot completed a formal, computer-certifiable proof of Fermat’s Last Theorem in 11 days, a project that had been estimated to require ten years of human effort. The accomplishment was hailed by mathematicians like Alex Kontorovich, who called the speed of the result astonishing, and Kevin Buzzard, who noted the rapid progress from earlier AI-assisted formalizations such as Maryna Viazovska’s sphere-packing work.

The original theorem, proved by Andrew Wiles and Richard Taylor in 1994, states that no three positive integers satisfy the equation xⁿ + yⁿ = zⁿ for n > 2. Claude’s success demonstrates AI’s growing capacity to both check existing proofs and generate new mathematical reasoning. Experts predict that, at one outlet pace, AI may eventually be able to audit the entire body of mathematical knowledge, potentially uncovering errors in long-standing results. The development marks a significant milestone in the intersection of artificial intelligence and pure mathematics.