Axiom Math AI Converts Stacks Project to Code

The AI startup's agents turned thousands of pages of algebraic geometry reference material into formally checked code.

Updated on Nov. 6, 2026 in Mathematics

Isometric editorial illustration of a tall stack of dark bound volumes beside a pale stone block joined by a thin line of light. — Axiom Math's AI agents converted about 7,654 pages of the Stacks Project into 16 million lines of formally verified code, exposing missing conditions and untrue claims. AI Illustration. Upload story photo >

Live Poll

Do you trust AI-generated work more when it has been computer-verified?

Axiom Math announced its AI agents converted about 7,654 pages of the Stacks Project into 16 million lines of code. The conversion uses the Lean system to verify each proof step follows logically from the previous ones.

Why it matters

The effort shows AI systems can formalize large bodies of mathematics that researchers would otherwise need to check by hand. The process also surfaced missing conditions and untrue claims in some original Stacks Project statements.

Axiom Math's AI agents converted about 7,654 pages of the Stacks Project into 16 million lines of code. The source does not say how many of the flagged statements have since been corrected in the original reference.

The players

Axiom Math

An AI startup focused on applying artificial intelligence to formal mathematics and verified software development.

Carina Hong

The founder of Axiom Math, who left a Stanford mathematics PhD program to start the company in 2025.

Johan de Jong

A mathematician who launched the Stacks Project in 2005 as a free online reference for algebraic geometry.

The details

Axiom Math's AI agents worked on thousands of mathematical problems at once while keeping the collection consistent. The system converts Stacks Project material into a form computers can examine step by step, and it uncovered missing conditions and untrue claims in some original statements.

2005: Johan de Jong launches the Stacks Project as a free online reference for algebraic geometry.
March 2025: Carina Hong leaves a Stanford mathematics PhD program to found Axiom Math.
January: Axiom's AI system solves all 12 Putnam competition problems.
March 2026: Axiom raises $200 million in a round led by Menlo Ventures.

Big Picture

The effort extends the Lean proof assistant from checking individual theorems toward formalizing an entire widely used reference text. It points to a trajectory in which AI helps verify mathematics at the scale of whole fields rather than single results.

Readers following math education or software verification may find the approach relevant to checking complex code. The company says it plans to apply the technology to finding errors in AI-generated software.

The takeaway

Verified formal libraries could make large bodies of mathematics easier to trust and build on. Readers tracking AI in science can watch whether such conversions hold up under independent review.