Bỏ qua đến nội dung chính
Back to home
AI tools-ai Tech 1 min read

Anthropic: Claude Successfully Formalizes Fermat's Last Theorem

Anthropic announced that its Claude model has completed a formal proof of Fermat's Last Theorem using the Lean interactive theorem prover.

Tier 1 · sources 99% confidence Reviewed
Sources x.com

A Milestone in Pure Mathematics

On September 4, 2026, Anthropic announced that its Claude model completed the first formalized proof of Fermat's Last Theorem over the past month, marking a significant milestone in applying artificial intelligence to pure mathematics. Manually verifying large-scale mathematical proofs traditionally demands years of labor from domain experts. Mathematical formalization—converting mathematical arguments into machine-checkable code for interactive theorem provers like Lean—is widely recognized as the most viable path to addressing this verification bottleneck.

Machine-Checked Rigor with Lean

According to Anthropic's announcement on X, Claude directly performed the translation and finalized the formal proof for the celebrated theorem. Interactive proof assistants like Lean require every logical step to be defined with absolute formal precision, completely removing the implicit assumptions or heuristic leaps typical of human-written mathematical literature. Once ingested into Lean, the system deterministically validates the entire logical chain without relying on subjective human judgment.

Advancing AI Reasoning Capabilities

Fermat's Last Theorem stood unsolved for more than three centuries before culminating in a multi-hundred-page proof. Translating a proof of such magnitude and mathematical depth into Lean's formal syntax represents a massive intellectual workload. This milestone demonstrates the advancing capabilities of frontier AI models in executing rigorous logical reasoning and interacting with formal verification environments.

Awaiting Full Technical Disclosures

Anthropic has not yet disclosed specifics regarding the total compute runtime, the volume of generated Lean code, or the precise model variant used in the experiment. The mathematics and computer science communities are actively anticipating the release of the complete formal repository and accompanying technical report for independent verification.