Anthropic has used its Claude AI system to produce a formal, machine-verified proof of Fermat's Last Theorem, one of the most celebrated results in the history of mathematics. The theorem, which states that no three positive integers can satisfy the equation a^n + b^n = c^n for any integer value of n greater than two, was first proved by mathematician Andrew Wiles in 1995 after more than 350 years as an open problem. Translating that proof into a form a computer can verify is a separate and deeply complex challenge, and one that Claude has now helped advance in a concrete way.
What Formalization Actually Means
A formal proof is not simply a restatement of existing mathematics. It requires every logical step to be expressed in a precise language that a proof assistant can check mechanically, leaving no room for the intuitive leaps that human mathematicians routinely rely on. Tools like Lean and Coq are commonly used for this purpose, and the process of encoding a proof this way can take years of expert effort even for results far simpler than Wiles's work on Fermat. Anthropic's use of Claude to formalize Fermat's Last Theorem represents an attempt to accelerate that process using a large language model capable of reasoning across lengthy, interdependent mathematical structures.
Key Facts
- Fermat's Last Theorem was first proved by Andrew Wiles in 1995 after more than three centuries as an unsolved problem.
- Formal verification requires encoding every proof step in a machine-checkable language such as Lean or Coq.
- Anthropic used Claude to assist in generating and checking the formal proof structure.
- The project is part of a broader push by AI labs to apply models to rigorous scientific and mathematical domains.
- Formal verification of Wiles's proof has long been considered one of the hardest open problems in computer-assisted mathematics.
The significance of this effort extends beyond the specific theorem. Mathematicians and computer scientists have long viewed the full formalization of Wiles's proof as a kind of benchmark for the maturity of formal methods. The proof draws on a vast range of modern mathematics, including elliptic curves, modular forms, and Galois representations, meaning any system capable of handling it must navigate extraordinary breadth and depth. Claude's involvement signals that large language models may now be capable of contributing meaningfully to tasks that were previously the exclusive domain of highly specialized human experts working over extended periods.
Formalizing Fermat's Last Theorem has been described by experts as one of the hardest challenges in computer-assisted mathematics, requiring fluency across enormous swaths of modern number theory and algebraic geometry.Mathematical community consensus, widely reported
AI and the Future of Mathematical Proof
This development fits into a pattern of Anthropic directing Claude toward high-stakes scientific domains. The company has been expanding its focus on research applications, including efforts in the life sciences where AI-assisted reasoning could accelerate drug discovery and experimental design. The move into formal mathematics is distinct but philosophically connected: in both cases, the goal is to apply AI to domains where correctness is non-negotiable and errors carry real consequences.
Formal verification has practical applications well beyond pure mathematics. Software systems, cryptographic protocols, and hardware designs can all benefit from machine-checked proofs of correctness. If AI models can help reduce the cost and time required to produce formal proofs, the downstream effects could touch a wide range of engineering and scientific fields. The Fermat formalization, while rooted in abstract number theory, acts as a proof of concept for that broader ambition.
It is worth noting that Claude's role in this effort is best understood as collaborative rather than autonomous. The model worked alongside human mathematicians and formal verification specialists, helping to generate candidate proof steps, identify gaps, and translate reasoning into the structured language required by proof assistants. The result is a human-AI collaboration of a kind that is likely to become more common as these tools mature. For those following Claude's expanding role in scientific research, this project adds another data point to a growing body of evidence that capable AI systems can contribute to the hardest problems in human knowledge.
The formalization effort is ongoing work in the field, and independent verification of the full result will be an important next step. Still, the progress reported here is a clear indicator that the boundary between AI-assisted mathematics and human-led mathematics is shifting, and shifting faster than many in the field anticipated.
“Formalizing Fermat's Last Theorem isn't just a mathematical milestone, it signals that AI can now operate as a genuine verification layer for complex reasoning, and organizations should start thinking seriously about where that capability fits into their knowledge workflows.”
Leon Tindemans, AI expert and entrepreneur specialising in Claude, Copilot and ChatGPT. Learn more with Copilot training by TTM Communicatie.
