42AI Emerges With $62.5M to Develop AI Model Predicting Human Behavior

Image: TechCrunch · Source

Israeli startup 42AI reveals a behavioral world model AI that predicts individual decisions probabilistically, securing $62.5 million seed funding and debuting in financial trading applications.

Israeli startup 42AI has come out of stealth mode, announcing it has raised $62.5 million in seed funding to develop an artificial intelligence system designed to predict human behavior in new and changing situations. The company claims its unique "behavioral world model" can forecast the probability of how an individual might act before they make a decision, even in circumstances they have never encountered before.

42AI's founders, Amit Gal-Or and Ehud Ezra, explain their model shifts focus from deterministic predictions to probabilistic behavioral tendencies rooted in psychological traits. Instead of simply extrapolating past actions, the AI analyzes a wide range of individual interactions—such as customer support calls, transaction history, onboarding procedures, and trading activity—to infer internal motivations and translate them into mathematical trait representations.

The system utilizes a transformer-based world model to simulate and assign probabilities to different decision outcomes depending on varying external conditions. Additionally, a dedicated language model generates explanations for its predictions, enhancing interpretability. The company emphasizes that while the model does not use personally identifiable information, it does create behavioral profiles from linked data to maintain accuracy.

Developed over three years with interdisciplinary insights from psychology, neuroscience, and genetics—and guided by experts including Nobel laureate Dan Shechtman—the model has been trained on datasets representing over 30 million individuals. Currently deployed in about ten large brokerage firms, 42AI’s model reportedly assists in managing liquidity and hedging strategies by predicting how traders will react to volatile markets.

In performance measures, the firm asserts a 98% accuracy rate in forecasting trading flows within expected ranges, with F1 scores between 0.8 and 0.85 for predicting trade direction and volume. However, they acknowledge limitations stemming from incomplete situational context, such as unrecorded external advice influencing decisions.

42AI’s business model involves charging clients a flat fee or taking a percentage of generated value; one brokerage reportedly gained approximately $30 million through dynamic hedging strategies informed by the AI. The company intends to expand beyond finance, exploring applications in gaming, fraud prevention, and cybersecurity, and eventually enabling robots and AI agents to better understand human behavior—though substantial domain-specific training will be necessary.

The founders also recognize ethical challenges, particularly the risk of their predictive technology being used by clients to unfairly advantage themselves over their own customers. They note the presence of regulatory oversight, monitoring, and an internal ethics committee as mitigations, but do not detail specific safeguards.

Despite these concerns, Amit Gal-Or expresses confidence that 42AI’s technology could be part of the solution to problems arising from opaque financial behaviors and unpredictable human decisions. The startup plans additional fundraising rounds to further develop and commercialize their AI model.