Anthropic’s Claude Completes First Formal Proof of Fermat’s Last Theorem in 11 Days

2 min read
Source: Decrypt
Anthropic’s Claude Completes First Formal Proof of Fermat’s Last Theorem in 11 Days
Photo: Decrypt
TL;DR

Anthropic’s Claude AI generated the first fully computer-checked proof of Fermat’s Last Theorem in 11 days, producing 13 million lines of code. This automated effort outpaced a human-led project at Imperial College London that has been running since 2024. The proof, verified by mathematician Kevin Buzzard, resolves a 358-year-old problem by translating Andrew Wiles’s 1995 solution into a machine-verifiable format.

Key points

  • Claude produced 13 million lines of code, equivalent to 160 novels, to formally verify Fermat’s Last Theorem.
  • The AI completed the task in 11 days, significantly faster than the human-led project at Imperial College London initiated in 2024.
  • Mathematician Kevin Buzzard confirmed the proof’s validity using basic logical axioms, noting it relies on no assumptions beyond standard mathematics.
  • The process utilized a tool called Prove2Me to coordinate dozens of parallel AI agents and prevent redundant work.
  • The proof is publicly available on GitHub, allowing independent verification by the mathematical community.

Background

This development follows earlier AI breakthroughs in mathematics, such as OpenAI’s Astra model solving ten long-standing problems in August 2026. It also builds on the historical context of Andrew Wiles’s 1995 proof, which took years to verify due to its complexity. The current event highlights a shift toward formal verification, where computers check every logical step, addressing concerns about unverified AI-generated proofs.

Why it matters

This milestone demonstrates AI’s capability to handle complex, long-term logical tasks with high precision, reducing the time required to verify major mathematical proofs from years to days. It addresses the growing backlog of unverified proofs in mathematics and provides a deterministic, error-free method for validation. The public availability of the proof on GitHub ensures transparency and allows the community to scrutinize the AI’s work, potentially accelerating the adoption of formal verification in academic research.

What to watch

Mathematicians will likely continue to scrutinize the 13-million-line proof on GitHub to ensure no logical gaps exist. The success of this project may prompt other institutions to adopt AI-assisted formalization tools for verifying existing proofs. Additionally, the development of tools like Prove2Me could lead to more efficient coordination of AI agents in other complex computational tasks.

Share this article

Want the full story? Read the original reporting

Read on Decrypt