AI
Axiom Math Completes Formal Verification of the Stacks Project to Build a Trusted Mathematical Foundation
Axiom Math, valued at $1600000000, announced that it has completed formal verification of the large mathematics knowledge base, the Stacks Project. Its AI agents converted roughly 7,654 pages into about 16000000 lines of computer-checked Lean code, aiming to provide a reliable foundation for AI agents writing software and eventually conducting scientific research.
Axiom Math hopes to avoid low-quality or unreliable results as AI produces large amounts of mathematical content. The company announced that it has formally verified the Stacks Project. The project was launched in 2005 by Columbia University mathematician Johan de Jong as an open reference resource, with content extending from college algebra to advanced ideas in algebra, geometry and number theory.
Formal mathematics involves rewriting proofs as code in the Lean language, which a proof assistant then checks line by line; once passed, the proof has greater logical reliability. Axiom’s AI agent team transformed the Stacks’ roughly 7,654 pages into about 16000000 lines of computer-checked code. The difficulty is that theorems and definitions in the library are interconnected, with every result built on preceding content, so all dependencies must remain consistent.
Company chief executive Carina Hong said the team initially was unsure whether this automated approach could preserve the structural beauty of Stacks. Technical staff member Ashvin Swaminathan said the system set up hierarchies but did not fully prescribe every agent’s task, and agents independently discovered work that could be handled in parallel. Axiom also trains its own mathematics models. The company was founded in March 2025 by Hong, now 25, who dropped out of Stanford’s mathematics PhD program; in March this year it completed a $200000000 Series A round led by Menlo Ventures.