Claude AI system interface displaying mathematical equations and computational results related to Riemann zeta function analysis

Anthropic’s Claude Formalizes Fermat’s Last Theorem, But Rival Proof Lands First

ByAlexey Bondarev September 5, 2026

Anthropic published research on Sept. 2 showing Claude models contributed to a machine-checked Lean formalization of Fermat’s Last Theorem, but mathematician Kevin Buzzard‘s team completed a full, sorry-free version first.

The timing makes this more than a routine research post. Fermat’s Last Theorem, proven by Andrew Wiles in 1994 after centuries of failed attempts, is one of the hardest results in number theory to translate into formal, machine-verifiable logic because the original proof spans hundreds of pages of advanced algebraic geometry.

Anthropic frames the work as evidence that large language models can now assist with genuinely difficult mathematical formalization, not just routine code generation, according to Anthropic. Buzzard’s team, however, reached a complete, gap-free Lean formalization before Anthropic’s effort closed every remaining piece, a detail highlighted by FourWeekMBA.

A Race Nobody Officially Entered

Neither side framed this as a competition at the outset, but the parallel timelines invite direct comparison between AI-assisted and human-led formalization speed. Buzzard has spent years building formalization infrastructure for major theorems, so his team’s head start reflects accumulated tooling as much as raw speed.

What The Gap Reveals

The episode exposes a real limit in current AI mathematical reasoning, Claude could contribute substantial pieces of the formalization but did not independently close the proof before human specialists did. That gap matters for anyone tracking whether AI models can autonomously verify frontier mathematics rather than assist alongside expert teams, and it’s worth asking what the formalization benchmark here actually measures about model capability versus scaffolding.

The result lands as Anthropic also faces political scrutiny, with Senator Bernie Sanders proposing a 50% public equity stake in major AI firms, a reminder that research milestones and regulatory pressure are now arriving in parallel.