Anthropic Claude AI Autonomously Formalizes Fermat Last Theorem Over 11 Days

Anthropic's Claude AI model autonomously formalized Fermat's Last Theorem in the Lean programming language over an 11-day period, researchers confirmed in early September 2026.

Fermat's Last Theorem, first proven by mathematician Andrew Wiles in 1995, states that no three positive integers can satisfy the equation a^n + b^n = c^n for any integer value of n greater than 2. Formalizing the proof in Lean means translating the mathematical argument into a machine-verifiable format that a computer can check step by step.

The task required Claude to work through thousands of logical steps without human guidance over the 11-day period. Researchers described it as a significant milestone in AI-assisted mathematics and a demonstration of the growing capability of AI systems to handle complex, long-horizon reasoning tasks.

The achievement comes as Anthropic also released Claude Fable 5.1, which reduces the cost of agentic workloads by up to 45 percent. The company has positioned itself as a safety-focused AI lab while continuing to push the boundaries of what its models can accomplish.

The broader AI landscape in early September 2026 is marked by rapid capability gains across multiple labs. OpenAI released GPT-6 Astra, which the company says marks the beginning of the AGI era. Microsoft released MAI-Transcribe-2 and unveiled Project Zenith for local AI development.

US lawmakers have responded to the pace of AI development by proposing the Ban Artificial Superintelligence Act, which would prohibit the development of superintelligent AI systems.