Claude Formalizes Fermat's Last Theorem, Beats Dedicated Academic Team
Anthropic's Claude helped complete a formal proof of Fermat's Last Theorem in Lean 4, beating Kevin Buzzard's team at Imperial College London who had been working on the same problem for years through the FLT project. Buzzard wrote a blog post titled 'Anthropic has beaten me to it' that is equal parts gracious and staggering. The thread notes this checks another box on Freek Wiedijk's list of 100 theorems that should be formalized.
The pattern here is not just AI-assisted math. It is AI doing the kind of sustained, technically precise reasoning that the field assumed would take decades to reach. Formal theorem proving requires not just cleverness but rigor, tracking proof state across thousands of steps without dropping a thread. Claude doing this in Lean 4 is a different category of result than generating code or summarizing documents.
The discussion is mostly awed rather than skeptical, which is itself a signal. When HN is not arguing, something genuinely surprising happened.
So what?
If AI can now do graduate-level mathematics with formal verification, the ceiling on what it can tackle in your codebase, your product spec, or your data model is higher than you priced in last month. Founders building tools that depend on 'AI can't reason about X' assumptions should re-examine those assumptions now.