Research

AxiomMath Formally Verifies Stacks Project, Building a Trusted Math Foundation for AI

The $1.6B startup has converted one of mathematics' largest reference libraries into machine-verifiable code, creating a logical foundation AI agents can rely on.

Last verified:

AxiomMath Formally Verifies the Stacks Project

According to TechCrunch AI, AxiomMath announced on November 5 that it has formally verified the Stacks Project, translating approximately 7,654 pages of one of mathematics’ most comprehensive reference libraries into roughly 16 million lines of machine-checkable code. The achievement represents a significant effort to ground AI-driven mathematical discovery in provably sound logic, rather than the “AI slop” that has proliferated across other domains.

The verification process converts mathematical proofs into Lean, a programming language designed for formal mathematics, and runs them through a proof assistant that validates each line’s logical correctness. This same technique is used by engineers to verify chip designs and has caught bugs in software for calculating government benefits in France. For a library as large and interconnected as Stacks, the challenge was preserving thousands of mathematical dependencies—theorems that reference earlier theorems—without introducing logical contradictions across 16 million lines of code.

Why AxiomMath Built on the Stacks Project

The Stacks Project began in 2005 when Columbia mathematician Johan de Jong established it as an open-source reference. Mathematicians have contributed continuously, building a resource that spans algebra, geometry, number theory, and advanced theoretical work. Columbia’s Peter Woit has described Stacks as having “nothing else really comparable” in mathematics, likening it to the Bourbaki books that fundamentally rebuilt mathematics from axiomatic foundations during the 20th century.

AxiomMath CEO Carina Hong, a 25-year-old who left a Stanford mathematics Ph.D. program to found the company in March 2025, emphasized the delicacy of the verification task. Speaking to TechCrunch, Hong noted: “You want [theorems] to be consistent and not, like, paradoxical. You want the definitions to not fight with each other.” The company’s approach—deploying AI agents in parallel with minimal constraints—allowed the team to preserve Stacks’ structural integrity across the formalization.

Funding and Strategic Vision

AxiomMath has attracted significant backing for its mission. In March 2026, Menlo Ventures led a $200 million Series A, bringing the startup’s valuation to $1.6 billion. The capital underscores investor conviction that formal verification of mathematical knowledge could become critical infrastructure for AI systems that generate proofs, write software, and eventually contribute to scientific discovery.

Why This Matters

The verification of Stacks creates a verifiable foundation that AI agents can reference when generating mathematical proofs or software with logical guarantees. As OpenAI alone released 719 manuscripts claiming mathematical solutions in recent months—some of which may contain errors undetected by automated checks—a formally verified library becomes increasingly valuable. Teams deploying AI to mathematics, chip design, or safety-critical systems now have a reference layer that has been machine-checked for consistency. This shifts the frontier from speed of discovery to trustworthiness of results, particularly relevant as AI agents take on roles in research, engineering, and policy domains where logical soundness is non-negotiable.

Frequently Asked Questions

What is formal verification and why does it matter for mathematics?

Formal verification rewrites proofs as executable code in a language like Lean, then uses a proof assistant to check each line for logical correctness. This catches errors that human review alone might miss, ensuring theorems are sound.

Why is the Stacks Project important?

The Stacks Project is a 7,654-page open-source reference covering algebra, geometry, and number theory. Mathematicians regard it as foundational—comparable to the Bourbaki books that rebuilt math from first principles in the 20th century.

How does AxiomMath's verification preserve the Stacks Project's interconnectedness?

The theorems in Stacks are interdependent; each builds on earlier results. AxiomMath's team used parallel AI agents with loose structure to preserve all ~16 million lines of dependencies across the 16 million lines of Lean code without introducing paradoxes or logical conflicts.

#formal-verification #mathematics #AI-agents #Lean #proof-assistant