An Israeli AI startup, 42AI, has emerged from stealth after raising $62.5 million in seed funding to power a “behavioral world model” aimed at predicting human decisions—even in unfamiliar situations. The model is being piloted at brokerages to forecast how traders might react during market volatility.
How 42AI’s Model Works
The company’s approach rests on the idea that while behavior isn’t deterministic, underlying psychological tendencies influence people’s choices across different contexts. Rather than relying solely on past actions, the model infers motivations by observing responses to diverse events—whether someone is onboarding, making support calls, transferring funds, or making trades. These observations are translated into mathematical traits, then plugged into a transformer-based world model that outputs probabilities for future decisions.
To explain its predictions, the system uses a separate language model. The training process rewards simulations that preserve consistent tendencies—for instance, treating someone who is usually risk-averse as risk-averse in the model’s predictions.
Data, Performance, and Early Use Cases
42AI built its foundation over three years, drawing on insight from psychological, neuroscience, and genetic research. Initial data came from psych assessments of thousands of people, later expanded with government and biobank datasets covering over 2 million individuals. Its customer data now reflects behaviors of nearly 30 million people.
Early deployments at around ten large brokerages are using the model for tasks like managing liquidity, margin, and hedging. The model boasts a 98% accuracy rate for forecasting whether aggregate trading flows fall within certain predicted ranges—not for predicting each individual trade. It also maintains an F1 score between 0.8 and 0.85 for predicting trading direction and volume.
Funding, Applications, and Risks
The $62.5 million seed round was partially funded by 42AI’s founders and Infinity Group—co-founded by the company’s chairman, who is also the CEO’s father—spread out over three years. The startup is already generating seven-figure revenue but has not disclosed full financials. One brokerage customer reported about $30 million in gained value through dynamic hedging driven by the model.
42AI says it doesn’t require personally identifiable information to make predictions, though it still builds individual behavior profiles by linking different user activities. Looking beyond finance, the company is exploring applications in gaming, fraud prevention, cybersecurity—and eventually enabling robots and AI agents to understand people better. But adding new domains will require fresh rules and context.
There are ethical considerations. If firms can anticipate what people will do, they could exploit that knowledge. The company says it has regulatory oversight and an internal ethics committee, but hasn’t detailed safeguards against misuse by its clients.
42AI acknowledges limitations: the model lacks full visibility into every influence on someone’s behavior—like external advice from friends or colleagues—leading to occasional mispredictions when it misses critical context.
Overall, 42AI is betting that human behavior can be understood not just from what people do, but from why they do it. By moving from pattern recognition to modeling internal motivations, the startup aims to redefine predictive AI.
What this means, and what to watch
Predicting behavior has always been a frontier in AI—and 42AI’s launch brings that frontier into sharper view. The company’s ability to blend psychological theory with advanced models could make its system unusually powerful in finance and beyond. But success will depend on managing bias, protecting privacy, and proving robustness across diverse domains. As 42AI raises its next round and begins applying its model in new areas, its ethical frameworks, regulatory compliance, and real-world accuracy will be under close scrutiny.
