42AI emerges with $62.5M and a model it says predicts how traders will behave

Plenty of startups claim to understand their customers. 42AI says it can forecast what a specific person will decide, even in situations that person hasn't faced before. The Israeli company launched publicly with a $62.5 million seed and what it calls a behavioral world model, TechCrunch's Ram Iyer reported.

How it's supposed to work

Rather than extrapolating from past trades, 42AI tries to infer stable tendencies, like how much risk someone tolerates, from many loosely related interactions. For a brokerage client that means onboarding, KYC checks, support calls (including voice and sentiment), money transfers and the trades themselves. Those signals become numeric trait representations fed to a transformer that outputs probabilities for possible decisions. A separate language model writes the explanations.

The founders say the work took three years and drew on psychology, neuroscience and genetics, with Nobel chemistry laureate Dan Shechtman among the advisers. Training started with psychological assessments of thousands of people, then added government and biobank data on more than 2 million. Customer datasets now cover nearly 30 million people, the company says.

Benchmarks vs. vibes

The big number is 98% accuracy, but CEO Amit Gal-Or told TechCrunch it's about whether overall trading flows land inside the range the model expected, not about getting individual trades right. The more meaningful figure is an F1 between 0.8 and 0.85 on trading direction and volume. The founders admit limits, too: the model can't see a tip from a friend, and each new domain needs its own rules and context.

Business claims: 10 large brokerages use it for liquidity, margin and hedging, revenue is "seven figures" with no exact number, and one brokerage reportedly gained about $30 million from dynamic hedging. The seed isn't a typical venture round either. It came over three years from the founders and Infinity Group, which was started by the CEO's father, who chairs 42AI and co-founded it.

The uncomfortable part

A broker that can predict what its clients will do can also use that against them. CTO Ehud Ezra called this a valid concern. The founders pointed to regulation, monitoring and an internal ethics committee, but not to anything that would actually stop a customer from doing it. And while they say the model never knows who a person is, it still builds individual profiles from linked activity, a distinction regulators may not find comforting.

Scoreboard update: AI +1. A model claims to read traders’ tendencies better than they might like.

Stay undead.