Israeli startup 42AI exits stealth with $62.5M for a model that predicts how traders behave

Israeli startup 42AI came out of stealth on 2 November 2026 with $62.5 million in seed funding. It is building what it calls a behavioral world model, and its first product forecasts how individual traders at brokerages will act when markets turn volatile.

For each client, the software puts odds on the choices that person might make, so a brokerage can see what that client is likely to trade in the next 24 hours: which instrument, in which direction and in what volume. Ten large brokerages already use it for liquidity planning, margin requirements and hedging exposure. Revenue is above seven figures, though no number was given, and pricing is either a flat fee or a share of the value created. One brokerage reportedly gained about $30 million through dynamic hedging.

“In a world model, every individual is the world,”

Ehud Ezra, co-founder and CTO of 42AI, to TechCrunch · techcrunch.com ↗

The seed money came over three years from the founders and Infinity Group, so it is not all outside capital, and the company plans to raise more. Infinity Group was founded by Amir Gal-Or, 42AI's chairman and co-founder and the father of CEO Amit Gal-Or. Nobel chemistry laureate Dan Shechtman advises the research team. CTO Ehud Ezra and three co-authors posted a paper in February 2026 that adapted Meta's 8-billion-parameter Llama 3.1 to predict strategic choices from a person's psychometric profile.

The model took three years to build. It was first trained on studies of thousands of people who also took psychological assessments, then on government and biobank data covering more than two million people, while customer data spans nearly 30 million. Personality traits are turned into mathematical representations and fed to a transformer-based world model, and a separate language model writes the explanations a user reads. Inputs include onboarding and know-your-customer records, support calls (how often, voice and sentiment), transfers and trades.

The headline accuracy claim is 98%, but that measures whether overall trading flows land inside a predicted range, not whether each trade is called right; for direction and volume the F1 score is 0.8 to 0.85. Accuracy falls when outside influences are invisible, such as a trader acting on a colleague's advice. The company says the system needs no personally identifiable information and never knows who a person is, although individual profiles are still built from linked account activity. Asked about misuse, Ezra called the concern valid and the founders pointed to regulation, monitoring and an internal ethics committee, without explaining what would stop a customer turning the tool against its own clients.

Gaming, fraud prevention and cybersecurity come next, and robots and AI agents later, with each new domain needing training on its own rules and context. The company's stated view is that behavior is probabilistic, and that people keep underlying tendencies even as their circumstances change.