ai
Start-ups hire mathematicians to build AI systems that verify their own math proofs
Several AI companies have raised hundreds of millions of dollars to develop systems that solve mathematics problems and verify the correctness of their outputs. The firms aim to use mathematical verification methods to reduce errors in AI-generated code and proofs.
flipboard.comAudio version
Tap play to generate a narrated version.
Developing·Limited corroboration so far. This page will refresh as more sources emerge.
AI start-ups are recruiting mathematicians and developing systems designed to solve advanced mathematics problems while producing results that can be checked for correctness. OpenAI is testing how its models perform on mathematical benchmarks. Other companies formed in the past two years, including Axiom Math and Harmonic, have secured hundreds of millions of dollars from investors to pursue similar goals.
Companies and funding Axiom Math is based in Palo Alto and was founded by Carina Hong, a former Stanford student. Ken Ono, previously a professor at the University of Virginia, joined the company on leave in 2025 after working with Epoch AI to create difficult math problems for testing AI.
Harmonic, located nearby, is developing what it calls a mathematical superintelligence that produces verifiable outputs. Both firms occupy modest offices but have received large investments.
Verification methods and revenue plans The companies plan to use formal verification tools such as the Lean programming language to confirm that AI-generated proofs and code contain no errors. Harmonic CEO Tudor Achim stated that verification demand will rise as AI produces more code.
Axiom Math has had five papers containing proofs generated by its AI tools accepted in mathematical journals. The company aims to produce dozens of such papers by next year.
firms and mathematician concerns OpenAI chief scientist Jakub Pachocki said mathematics serves as a measurable benchmark for AI progress. Sébastien Bubeck at OpenAI noted that recent models have moved beyond earlier limitations in handling mathematical arguments.
