42AI Exits Stealth With $62.5 Million for a Behavioral World Model That Scores 0.8 to 0.85 F1 on Trader Calls

42AI, an Israeli startup that has spent three years building what it calls a "behavioral world model," came out of stealth on Monday with $62.5 million in seed funding and a claim that borders on science fiction: that its system can estimate what a specific person will do in a situation it has never observed them in.

The money is not a conventional venture round. Chief executive and co-founder Amit Gal-Or told TechCrunch that the $62.5 million came from the founders and Infinity Group over three years and was not entirely raised from outside investors. Infinity was founded by his father, Amir Gal-Or, who is also 42AI's chairman and a co-founder.

The mechanism is deliberately indirect. Rather than extrapolating from a person's past actions, 42AI studies how they behave in adjacent, seemingly unrelated moments — finishing a know-your-customer check, sitting through onboarding, calling support, moving money, placing a trade. Sentiment, voice characteristics and how often they call all go in. Those observations are converted into mathematical representations of traits and fed into a transformer-based world model that outputs probabilities for decisions under different conditions. A separate language model explains the output.

The training objective is built to keep a person's underlying tendencies stable across simulations: if someone who normally avoids risk is suddenly treated as a risk-taker inside a scenario, the target is treated as off.

The scale claims are large and entirely company-reported. Gal-Or says the team began with psych assessments and behavioral scenarios covering thousands of people, then added government and biobank datasets covering more than 2 million, and that customer datasets now reach almost 30 million people. He also says the system does not require personally identifiable information — "the model never knows who the person is" — although the company still builds individual behavioral profiles from linked customer activity.

In live use, 42AI says 10 large brokerages run the model to manage liquidity, margin requirements and hedging exposure. It claims 98% accuracy on market trades, a figure Gal-Or narrowed almost immediately: the metric measures whether trading flows land inside a predicted range, not whether each individual trade was called correctly. For direction and volume he cites an F1 score of 0.8 to 0.85. Revenue is "seven figures," he said, without giving a number. One brokerage is said to have generated roughly $30 million in additional value through dynamic hedging. None of these figures has been independently audited.

Gal-Or is also candid about the ceiling. "There's a natural ceiling because you really don't have access to everything that a person experiences," he said. Accuracy degrades when situational context is missing — a trader acting on a colleague's tip, for instance. Extending the model into gaming, fraud prevention, cybersecurity or robotics would require retraining rules and context for each new domain.

The obvious question is what stops a brokerage that can anticipate its clients' next move from trading against them. Co-founder and CTO Ehud Ezra called it "a valid concern"; the founders point to regulation, monitoring and an internal ethics committee, but did not describe any technical safeguard. Gal-Or's own answer was that the company is "part of the solution." For a product whose value rests precisely on asymmetric information about individuals, that will need considerably more than an ethics committee behind it.