42AI Emerges From Stealth With $62.5M to Predict Individual Trader Behavior

Israeli startup 42AI has come out of stealth with $62.5 million in seed funding for what it calls a behavioral world model, a system built to predict human behavior. The TechCrunch report, published on November 2, 2026, says the company's first use case is predicting how individual traders respond to volatile conditions.

CEO and co-founder Amit Gal-Or and CTO and co-founder Ehud Ezra lead the company. According to the same report, 42AI is running its system at a handful of financial brokerages. The deployment is at an early stage, and the stated goal is per-trader behavioral forecasting rather than market-level signal.

The architecture, as the company describes it, has three parts. Observations about a person are converted into mathematical representations of traits, essentially vectors of numbers that stand in for characteristics. Those representations feed a transformer-based world model, a neural network architecture that outputs probabilities across possible decisions. A separate language model then generates the explanation attached to each prediction. 42AI says its explainability layer identifies which traits, context and interactions were triggered in a given prediction.

42AI says it spent three years building the approach, combining psychological research with neuroscience and genetics. The training base started with thousands of people who completed psychological assessments and behavioral scenarios. The company then added government and biobank datasets covering more than 2 million people, and says its customer datasets now cover almost 30 million people.

The headline performance claim is 98% accuracy in predicting market trades. The company defines accuracy as whether trading flows fall within a predicted range, not whether each individual trade was predicted correctly. Those are different tests, and a range-based measure is more forgiving than per-trade classification. Reported latency is 1 to 10 milliseconds for the system that predicts behavior across time, and 2 to 12 milliseconds for the system that predicts responses to specific events.

In my view, the accuracy definition is the first number an expert reader should examine. A forecast that flow lands inside a band can help a desk managing risk without forecasting any particular order. Whether the band is tight enough to matter operationally is a question the headline figure does not answer on its own.

The broader context here is the data. Combining psychological assessments, genetic and biobank information and commercial customer records is a different proposition from modeling market microstructure. Predicting how a named individual will react under stress raises questions about consent, the purpose for which the data was collected, and where else the same models might be applied. Those questions stand whether or not the trading use case performs as claimed.

The other side of the argument deserves a hearing. An explainability layer that reports which traits and contexts drove a prediction could, if it holds up under scrutiny, give supervisors and clients a more legible account of automated decisions than a bare probability does. Whether that transparency offsets the sensitivity of the underlying inputs is a judgment the coming deployments will test.

For now, 42AI has a funding round, a stated architecture, a small set of brokerage customers, and performance claims governed by its own definitions. The next useful evidence will come from how those claims hold up once more clients run the system against live volatility.