A startup called 42AI says it is building a model that estimates how people may act in situations they have not encountered before. The company has emerged from stealth with $62.5 million in seed funding and is applying the technology at financial brokerages, where it aims to forecast how traders might respond to volatile conditions.
Building a picture from many interactions
42AI describes its system as a behavioral world model. Its premise is that people’s choices are uncertain, but underlying tendencies can still influence decisions across different circumstances. The model aims to estimate those tendencies and express possible decisions as probabilities.
Rather than simply projecting a person’s past actions forward, the company says it looks for clues about their motivations in a wider range of interactions. In finance, those inputs can include identity checks, onboarding, support calls, money transfers and stock trades.
The system may examine the sentiment and voice in conversations, how often someone contacts support, and later details about when and how they buy or sell. The company says it combines these observations into mathematical representations of traits, then uses a transformer-based model to estimate possible choices in different situations. A separate language model generates explanations for the predictions.
That approach depends on information from more than one moment or type of activity. A support call on its own may say little about a later trade, but 42AI’s method treats many such signals as parts of an individual behavioral profile. The company says it does not need personally identifiable information, while the article notes that it appears to create profiles from linked customer activity.
What the company says its results show
42AI says it spent three years developing the approach, drawing on psychological research, neuroscience and genetics, and advice from experts in psychology and PTSD, as well as Dan Shechtman, who was awarded the 2011 Nobel Prize in Chemistry.
According to CEO and co-founder Amit Gal-Or, the team initially studied thousands of people through psychological assessments and behavioral scenarios. It later added government and biobank datasets covering more than 2 million people. Customer datasets have since grown to almost 30 million people, he said.
For market trading, 42AI claims 98% accuracy. Gal-Or clarified that this figure measures whether trading flows fall within a predicted range; it does not mean the model correctly forecasts every individual trade. He also said the system reaches an F1 score of 0.8 to 0.85 when predicting trading direction and volume. F1 is a machine-learning measure that balances correct classifications against false alarms and missed events.
The model is intended to preserve traits through simulations. For instance, if a person is usually risk-averse, a simulation that suddenly treats them as a risk-taker would count as a miss, CTO and co-founder Ehud Ezra said. That design makes consistent representation of behavior central to the system’s claimed performance.
Limits, customers and potential uses
The model cannot observe everything that shapes a person’s choices. Gal-Or said there is a natural ceiling when a system lacks access to everything someone experiences. Ezra gave the example of a trader changing decisions after advice from a friend or colleague: if the model lacks that situational context, its accuracy can suffer.
Gal-Or said 42AI is working with 10 large brokerages, which use the system to manage liquidity, margin requirements and hedging exposure. He said the company is onboarding more customers and generating more than “seven figures” in revenue, without specifying the amount.
Customers pay either a flat fee or a percentage of the value the platform helps generate. Gal-Or claimed that one brokerage gained approximately $30 million in additional value through dynamic hedging. The company’s $62.5 million seed funding came from its founders and Infinity Group over three years, and was not entirely raised from external investors. Infinity was founded by Gal-Or’s father, Amir Gal-Or, who is also 42AI’s chairman and a co-founder. The startup now plans to raise more money.
42AI is also exploring gaming, fraud prevention and cybersecurity. Its longer-term ambition is to help robots and AI agents understand people, though Ezra said the model would need training on the rules and context of each new domain.
Who benefits from behavioral forecasts?
Predicting customer behavior could help a brokerage plan for changing trading activity. The same capability also raises a question: if a firm can anticipate what customers may do, how will it keep that information from being used to the firm’s advantage at the customer’s expense?
Ezra called the concern “a valid concern.” The founders pointed to regulations, monitoring and an internal ethics committee, but did not explain what would prevent customers from using the technology to gain an advantage over the people it profiles. Gal-Or said the company believes its technology can help address the problem.
For now, 42AI’s claims rest on its ability to infer stable tendencies from incomplete and varied information. Its reported metrics describe performance on particular prediction tasks, while the company itself recognizes that missing context can change behavior. How brokerages use these forecasts is therefore part of the story, alongside how accurately the model makes them.