AI
Axiom Math formally verifies the Stacks Project to build foundations for AI
Axiom Math says it has formally verified the Stacks Project, a large open-source mathematics reference, by translating roughly 7,654 pages into about 16 million lines of Lean code checked by proof-assistant software. The company used parallel AI agents, a loose task hierarchy and math-specialized models to preserve connections among the project’s interdependent results. Axiom argues that verified mathematical libraries could give future AI agents reusable foundations for writing software and conducting research, though that broader transfer remains its thesis rather than a demonstrated outcome.
Editorial summary · Publisher article
What matters
- The Stacks Project spans algebra, geometry and number theory and was built by many contributors.
- Proof assistants check formalized proofs line by line for logical consistency.
- Axiom says its agents worked in parallel while maintaining dependencies across the library.
- The company sees formalized mathematics as infrastructure for more reliable AI-generated code.
Why formal verification matters
Formalization rewrites mathematical proofs as code that a proof assistant can check. The same broad approach is used in engineering and software research to detect flaws, but applying it to Stacks required preserving millions of interconnected dependencies.
AI agents as maintainers of a mathematical foundation
Axiom gave its agents a hierarchy without prescribing every task, allowing them to identify parallel work. The company also trained models for mathematics and says repeated exposure helped them develop a sense of which formalizations were useful and elegant.
From checked proofs to future software
Axiom’s central argument is that each verified result can become a reusable building block for later proofs and agent-written infrastructure. The proposed benefit is reduced propagation of errors, but the article presents this as an expected transfer from mathematics to software rather than an independently established result.
Context & caveats
- The article reports Axiom’s account of the project and its future applications; it does not independently test the company’s broader claims about AI-generated software.
Why this made the edition
Substantial reporting explains a large formalization project, its verification method, technical workflow and implications for mathematics and AI-written software, with company claims attributed.
Source material