AI proved a 300-year-old math challenge in just 11 days?

Imagine you challenge yourself to solve an ancient cipher that no one has cracked for 300 years. Someone else might be pursuing research for five long years, funded by government grants. How would you feel if an AI suddenly appeared and announced it had solved that cipher in just 11 days?

The recent announcement from AI company Anthropic is exactly that kind of situation. They claim their AI model, ‘Claude’, has completely proven a massive mathematical challenge, ‘Fermat’s Last Theorem (FLT)’, in a form that can be verified by a computer [References 1, 13, 15].

Why is this news important?

Asking an AI in your daily life to ‘organize my emails’ is a completely different story from having it solve a mathematical challenge. This is because mathematical proof is a rigorous process that allows not a single line of logical error.

The fact that AI has moved beyond simply writing well to independently constructing and verifying complex mathematical logic means that the way AI ‘thinks’ has evolved a step further. It is a powerful signal that we are entering an era where AI can assist us much faster and more accurately in complex scientific calculations and logical problem-solving that we previously found difficult [Reference 15].

Understanding the basics: Mathematical proof and ‘Lean’

For a moment, what is a ‘proof’ in the eyes of mathematicians? Simply put, it is the process of creating a ‘perfect instruction manual that no one can refute.’ Previously, humans reviewed it by writing on paper, but now, it is an era where one writes it in a language a computer can understand and receives mechanical confirmation that there are no errors. The tool used here is a programming language called ‘Lean’ (a tool that allows computers to verify mathematical proofs) [References 1, 3, 13].

To use an analogy, ‘Fermat’s Last Theorem’ is like putting together a puzzle with thousands of pieces. Mathematicians have been drawing the big picture of that puzzle, and Claude has been filling in the puzzle pieces according to that picture at a very high speed [References 4, 13].

Claude’s achievement reminds us once again how difficult and long the process of mathematical proof is. Professor Kevin Buzzard had even received a five-year research grant to formalize this theorem, but Claude performed this task ‘largely autonomously’ in 11 days [References 4, 13].

Current situation: A perfect conquest or joint research?

However, the view of the mathematics community on this news is somewhat complex. Anthropic announced that Claude solved this problem independently, but the math community says it is difficult to see this result as ‘AI doing it all by itself’ [References 7, 12].

This is because mathematicians around the world were already collaborating for years to assemble this puzzle, and Claude’s achievement is significantly built upon that foundation. In fact, the formalization of Fermat’s Last Theorem is still an ongoing community joint project, and there are loud voices that this result should not be considered the only first formalization proof [Reference 7].

In other words, AI is like an athlete who ran at incredible speed over a bridge of knowledge built by human mathematicians over hundreds of years. The speed was much faster, but humans are still the ones who built the bridge itself [References 9, 10].

What happens next?

The encounter between AI and mathematics will become even more intense in the future. This result has been released under the Apache License 2.0, making it accessible to everyone [Reference 5]. AI has now begun to show its potential not just as an assistant summarizing problems, but as a ‘co-researcher’ solving scientific challenges that humanity has not yet solved.

We will now achieve complex scientific discoveries much faster with AI. The day is not far off when AI, as an intellectual tool for humanity, will become a compass to explore territories of knowledge that humanity has not yet reached.

MindTickleBytes AI Reporter’s Perspective

While it is clear that AI has innovatively increased the speed of solving challenges, the essence of mathematics lies in the process of finding the answer itself. This case shows that AI can be a powerful tool that expands human intellectual boundaries. We look forward to seeing how AI and human mathematicians will cooperate in the future to open up a wider world of knowledge.

References

1. [FLT: Anthropic has beaten me to it

   Xena](https://xenaproject.wordpress.com/2026/09/04/flt-anthropic-has-beaten-me-to-it/)

2. [Formalizing Fermat’s Last Theorem

   Anthropic](https://www.anthropic.com/research/formalizing-fermats-last-theorem)

3. Techmeme: Anthropic says Claude worked “largely autonomously”…
4. GitHub - anthropics/fermats-last-theorem · GitHub
5. Fermat’s Last Theorem – from Wolfram MathWorld
6. Fermat’s Last Theorem in Lean: The Community… - DEV Community
7. ‘Amazing’ Math Bridge Extended Beyond Fermat’s Last Theorem
8. Proving Fermat’s last theorem: 2 mathematicians explain how building…
9. Fermat’s Last Theorem - Wikipedia
10. Claude helps complete first formalized proof of Fermat’s Last Theorem

11. [Learning more about Claude’s mathematical capabilities

    Anthropic](https://www.anthropic.com/research/riemann-zeta)

12. Lisan al Gaib on X: “Anthropic just uploaded a Lean 4 proof for Fermat’s last Theorem”
13. Anthropic Says Claude Produced Full Proof of Fermat’s Last Theorem, Verified With Lean