The archive · AI & Models · Product decision · 2025–2026
Axiom's verified-AI bet: perfect Putnam 12/12, $200M at $1.6B for provably safe code
Carina Hong's Axiom trains AI to prove rather than guess — Putnam 12/12 in Dec 2025, then a $200M Series A to verify AI-generated code in Lean.
Axiom Math (Axiom Quant Inc.)
What the business is
Axiom builds AxiomProver, an AI that produces machine-checkable mathematical proofs in Lean and applies that rigor to verify AI-generated code, with the stated endgame of an AI mathematician that proposes and proves new conjectures.
Starting capital：$64M seed (Oct 2025); $200M Series A led by Menlo Ventures with Greycroft and Madrona at a $1.6B valuation (Mar 2026).
How it started
Carina Hong, a Guangzhou-born MIT graduate and Stanford PhD dropout with the Morgan and Schafer prizes, founded Axiom in Palo Alto in 2025 with mathematician Ken Ono, ex-FAIR and Nvidia engineer Shubho Sengupta, and François Charton, who first applied transformers to a 130-year-old math problem; the company raised a $64M seed in October 2025.
What happened
In December 2025, AxiomProver scored a perfect 12/12 on the Putnam, solving 8 of 12 within exam time, and later verifiably proved a 20-year-old number-theory conjecture that Ono had never solved. On March 12, 2026, Axiom announced a $200M Series A led by Menlo Ventures with Greycroft and Madrona at a $1.6B valuation, to scale training infrastructure and make formal verification fast and affordable.
No ending yet — it is still running.
Background
Axiom is a Palo Alto startup founded in 2025 by Carina Hong, a 25-year-old MIT graduate and Stanford PhD dropout with the Morgan and Schafer prizes, together with mathematician Ken Ono, former Facebook AI and Nvidia engineer Shubho Sengupta, and François Charton, who first applied transformers to a math problem that stumped experts for 130 years. It raised a $64M seed in October 2025.
The product, AxiomProver, is trained to produce formally verified outputs in Lean, a programming language designed for mathematical proofs. Deterministic proof verifiers check each step, so the company claims mathematical certainty that generated code returns the right answer and does not introduce hidden vulnerabilities — a direct answer to the 'frequently works is a terrifying standard' problem Menlo Ventures' partners say plagues probabilistic AI code.
In December 2025, AxiomProver scored a perfect 12/12 on the Putnam, the most prestigious undergraduate math exam in North America, solving 8 of 12 within exam time; only five humans have achieved a perfect score in a century. It then verifiably proved a 20-year-old open number-theory conjecture that Ono had repeatedly failed to solve.
On March 12, 2026, Axiom announced a $200M Series A led by Menlo Ventures with Greycroft and Madrona at a $1.6B valuation, to scale training and expand its team of math experts. The company frames verification as a performance advantage: proof-checked outputs feed back into training as a 'verified data flywheel' that avoids the model collapse that plagues unverified AI.
What has to be true
- The bet is a standards argument, not a features argument: only provably correct AI code is safe enough for critical infrastructure.
- The Putnam 12/12 and the proved 20-year-old conjecture are concrete, checkable proof points that most AI labs cannot match.
- The verified data flywheel gives the company a defensible moat: every proof adds to training data without the risk of model collapse.
- It is a distinct company, event and year from every library entry — no other case covers formal verification of AI-generated code.
- The founding team combines prize-winning mathematics, FAIR/Nvidia systems engineering and transformer-era ML research.
What can be applied
Sell the arbiter, not the tool: Axiom doesn't claim better code, it claims provably correct code — a standard that turns 'hallucinations' from a bug into an architectural reason to buy.
Aftermath
As of September 2, 2026, Axiom continues to operate from Palo Alto with a roughly 20-person team and has not disclosed commercial customers or revenue; its public proof points remain the Putnam 12/12 and the proved conjecture. The company says it is scaling training infrastructure and expanding its team to make formal verification fast and affordable for every company shipping AI-generated code.
Sources
- Verifiable AI startup Axiom raises $200M to prove AI-generated code is safe to use
- Axiom Math raises $200M to verify AI-generated code with mathematics
- Fundraising
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card