Anthropic has achieved a staggering milestone in automated reasoning, using a prototype of its AI chatbot Claude to formalize the proof of Fermat’s Last Theorem. Announced on September 4, the breakthrough involved translating Andrew Wiles’ landmark 1994 work into a 13-million-line, computer-verifiable proof. The AI completed this massive undertaking in just 11 days, a task originally projected to take human mathematicians a decade to accomplish.

The Scale of the Achievement

For those unfamiliar with the depth of this problem, Fermat’s Last Theorem states that no three positive integers x, y, and z satisfy the equation x^n + y^n = z^n for any integer value of n greater than 2. While the statement is deceptively simple, the proof requires intricate connections between elliptic curves and modular forms. Alex Kontorovich, a number theorist at Rutgers University, stated that the machine’s ability to handle this complexity “just completely blew my mind.” The result is not merely a calculation but a rigorous formalization in the Lean programming language, ensuring every logical step is machine-checked.

From Fantasy to Fact

This development marks a sharp acceleration in AI’s mathematical capabilities. In February, AI had already certified Maryna Viazovska’s Fields-medal-winning work on sphere packing, but Kevin Buzzard of Imperial College London notes that the Fermat proof was “maybe an order of magnitude more difficult.” Buzzard admits that two years ago, the idea of AI scrutinizing the entire library of mathematical knowledge was “a fantasy.” Now, with Claude handling such dense theoretical frameworks, the possibility of discovering errors in established theorems is becoming a tangible reality.

The Bottom Line

If Anthropic can formalize Fermat’s Last Theorem in under two weeks, the era of purely human-verified mathematics is effectively over; AI is no longer just a calculator, but a peer reviewer capable of digesting centuries of abstract thought in days.

Key Takeaways

  • Anthropic’s prototype Claude formalized Andrew Wiles’ 1994 proof of Fermat’s Last Theorem in 11 days.
  • The resulting proof consists of 13 million lines of code, verified using the Lean programming language.
  • Mathematicians like Alex Kontorovich and Kevin Buzzard describe the achievement as unprecedented and mind-blowing.
  • This progress suggests AI may soon audit the entire existing body of mathematical knowledge for errors.