42AI comes out of stealth with $62.5M and a world model that can ‘predict human behavior’

Predicting Behavior with AI

Israeli startup 42AI has emerged from stealth with $62.5 million in seed funding to commercialize its “behavioral world model,” a system it claims can predict how individuals will react in entirely new situations. The company is initially deploying its technology with financial brokerages to anticipate trader responses to market volatility. By modeling what it describes as underlying psychological tendencies rather than just past actions, 42AI aims to provide a more nuanced layer of predictive analytics for risk management and financial planning.

How the Behavioral Model Works

The core of 42AI's technology is a transformer-based world model that processes a wide array of data points to build a probabilistic profile of an individual's decision-making framework. Co-founder and CTO Ehud Ezra explained that the system analyzes everything from sentiment in support calls to the speed of KYC processes alongside direct trading activity. The company reports achieving a consistent F1 score between 0.8 and 0.85 for predicting trading direction and volume in its current deployments.

• Data Inputs: The model ingests customer records including support calls (voice, sentiment), onboarding processes, money transfers, and transaction history.
• Modeling Approach: It focuses on inferring internal motivations and psychological tendencies, not just pattern-matching previous actions, to predict behavior in novel scenarios.
• Technical Core: A transformer-based world model converts observed traits into mathematical representations, then simulates outcomes and outputs probabilities for specific decisions.
• Funding Details: The $62.5 million seed round was sourced over three years from the founders and Infinity Group, a firm founded by the CEO's father.

Market Impact and Ethical Questions

Currently, 42AI is working with 10 large brokerages to help them manage liquidity and hedging exposure, with one client reportedly generating $30 million in additional value via dynamic hedging. The startup is planning to expand its model's application to gaming, fraud prevention, and cybersecurity. However, the technology raises significant ethical questions about the potential for firms to use these predictions against their own customers. The founders acknowledged this was a “valid concern” and pointed to regulations and an internal ethics committee as safeguards, without detailing specific mechanisms to prevent client misuse.

While 42AI's behavioral model presents a novel approach by focusing on psychological traits over historical actions, its real-world efficacy and ethical guardrails will face intense scrutiny as it expands beyond the controlled environment of financial trading.