Brokerages already know what their clients own. The expensive surprise is what those clients are about to do. 42AI launched today with $62.5M and a model built to forecast that second part.
Behavior as a probability, not a fact
The Israeli startup encodes what it has learned about an individual as numbers standing in for personality traits, and a transformer-based world model does the probability work. A second model, this one linguistic, spells out the traits and circumstances behind each forecast. For a brokerage, that translates into a 24-hour view of what each client is likely to trade, in which instrument, at what size and in which direction, so a risk desk can hedge before the exposure arrives.
Where the data comes from
The modeling effort behind it ran three years. Its first trial gathered psychometric answers from thousands of volunteers, and before work shifted to biobank and government archives the count had climbed beyond 2 million. Profiles drawn from clients cover some 30 million individuals. Ten large brokerages are paying clients, on flat fees or a cut of value generated, and revenue has reached seven figures, chief executive Amit Gal-Or said. By the founders’ reckoning, hedging dynamically delivered roughly $30M of extra value to a single brokerage.
The uncomfortable questions
The obvious one is whether a brokerage could aim those predictions at its own clients. Chief technology officer Ehud Ezra concedes the point, and lists regulation, monitoring and an internal ethics committee as guardrails. Gal-Or says identifiers are never needed and the model never learns who anyone is, though profiles are still assembled for every client.
Next up are gaming, fraud prevention and cybersecurity. Teaching robots and AI agents to read people sits further out.
