1 hour ago
17
A startup founded by a Stanford dropout less than 15 months ago claims to have done something that typically takes human mathematicians decades: solve multiple unsolved problems in mathematics using an AI system, then get the work accepted by the academic establishment.
Axiom Math, a Palo Alto-based company led by founder Carina Hong, says its AxiomProver system has cracked at least four previously unsolved mathematical problems. The proofs were posted on arXiv in February 2026, but as of late May 2026, no peer-reviewed journal publications have been confirmed — a distinction the article’s framing obscures.
What Axiom actually solved
The problems tackled by AxiomProver aren’t trivial exercises. Among the results are solutions to significant conjectures in algebraic geometry and Fel’s Conjecture, a problem connected to the work of Srinivasa Ramanujan. One of the solved problems involves a 20-year-old number theory conjecture.
At least one of the proofs resulted from collaboration with established mathematicians, not pure machine output.
The company uses formal verification through the Lean proof assistant. Rather than generating text that looks like a proof and hoping it’s correct, AxiomProver produces proofs that are mechanically checked for logical validity at every step. This formal verification layer is what gives the work credibility that typical large language model outputs don’t have.
The money and the team behind it
Axiom Math raised $200 million in Series A funding led by Menlo Ventures, pushing its post-money valuation to $1.6 billion. That came on top of $64 million in earlier seed investments, bringing total known funding to $264 million. Hong left a joint J.D./Ph.D. program at Stanford to start the venture in March 2025.
Ken Ono, a prominent number theorist known for his work on Ramanujan’s legacy and modular forms, joined Axiom Math as a founding mathematician.
The bigger picture for AI and mathematics
Axiom Math isn’t operating in a vacuum. Google’s DeepMind has been pushing into mathematical reasoning, most notably with its AlphaProof system.
It’s standard practice for mathematicians to post preprints on arXiv while papers undergo peer review, a process that can take months or even years. The February 2026 arXiv postings are currently undergoing expert scrutiny, and no peer-reviewed publications have been confirmed as of late May 2026.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·