Israeli startup 42AI has emerged from stealth with $62.5 million in seed funding for what it calls a "behavioral world model" — software that estimates not just what a person might do next, but why. The company, which spent three years building in private, is already running its system at 10 large financial brokerages, where it predicts how individual traders will respond when markets turn volatile.
Chief executive and co-founder Amit Gal-Or told TechCrunch, which first reported the launch on November 2, that behavior should be treated as probabilistic rather than fixed. "Behavior is not deterministic; it's constantly probabilistic," he said. The idea: underlying psychological tendencies persist as circumstances change, so a model that understands them can anticipate decisions in unfamiliar situations.
How the model claims to work
42AI says it starts by turning what it knows about a person into mathematical representations of personality traits, which a transformer-based world model then uses to work out the probabilities of different choices. A separate language model writes up which traits and context drove each prediction, so predictions come with an explanation.
For brokerage customers, the inputs include onboarding and know-your-customer records, support and sales interactions, money transfers and trades. Signals include sentiment, voice, call frequency, trade timing and market conditions. The company says a brokerage can see what each client is likely to trade over the next 24 hours, including the instrument, the direction and the volume.
The research began with thousands of people studied through psychological assessments and behavioral scenarios, before expanding to government and biobank datasets covering more than 2 million people. Customer datasets have since grown to cover almost 30 million people, Gal-Or said. A February paper from Ezra and co-authors described adapting an 8 billion-parameter Meta Llama 3.1 model to predict strategic choices from a psychometric profile.
The 98% claim — and its caveat
42AI's boldest number is a claim of 98% accuracy in predicting market trades. Gal-Or was careful to qualify it: the figure measures whether trading flows land inside a predicted range, not whether every individual trade is called correctly. He also reported F1 scores of 0.8 to 0.85 for predicting trading direction and volume.
The brokerages use the system to manage liquidity, margin requirements and hedging exposure, and 42AI charges either a flat fee or a percentage of the value it helps generate. By the company's account, one brokerage gained about $30 million in additional value through dynamic hedging, and revenue has reached seven figures.
The privacy question it cannot dodge
Gal-Or's answer to TechCrunch was blunt: the model does not need personally identifiable information, and "the model never knows who the person is." Profiles are still built for each individual from linked account activity, however, and Ezra acknowledged that a brokerage turning predictions against its own clients is "a valid concern." The founders pointed to regulation, monitoring and an internal ethics committee as checks on that kind of use.
The $62.5 million was not raised in a splashy external round. It came from the founders themselves and from Infinity Group over the three years of stealth building; Infinity was founded by Amir Gal-Or, who is 42AI's chairman and a co-founder. It plans to raise more, and is exploring gaming, fraud prevention and cybersecurity next, with robots and AI agents understanding people as a longer-term goal. It enters a crowded field: World Labs raised $1 billion, Yann LeCun's AMI Labs roughly €890 million, and Simile $100 million for a similar niche. 42AI's bet is that behavior itself — not language, not images — is the next thing a world model should learn.
