Axiom Math used AI agents to verify the Stacks Project library

Startup Axiom Math has introduced a system of AI agents designed to automatically verify mathematical proofs. The developers have successfully completed the formal verification of the Stacks Project — one of the world's largest algebraic geometry libraries, containing thousands of pages of theoretical derivations.

The company's technology aims to solve the problem of "AI slop" generated by modern language models. Unlike standard LLMs, Axiom Math's agents use formal verification methods that allow for the mathematical confirmation of every step in a proof. This eliminates the possibility of logical errors or hallucinations typical of probabilistic models.

The use of formal methods in mathematics has traditionally required a huge time investment from scientists, as writing code for proof assistant systems (such as Lean or Coq) is a labor-intensive process. The startup's agents automate this stage by translating natural mathematical language into a machine-readable format suitable for rigorous verification.

For the mathematical community, this means the ability to quickly confirm new hypotheses and check existing work for hidden errors. The implementation of such tools allows researchers to leverage the power of AI to accelerate discoveries while maintaining academic rigor and the reliability of results.