Israeli startup 42AI has come out of stealth with $62.5 million in seed funding for a behavioral world model that, it says, can estimate how people will act before they decide, even in situations they have never faced. The system is already running at 10 large brokerages, where it is used to manage liquidity, margin requirements, and hedging exposure in volatile markets.
What 42AI says it built
42AI argues that behavior is probabilistic rather than fixed, and that people carry underlying psychological tendencies that shape their choices even as circumstances change. Its model tries to measure those tendencies and convert them into data, rather than simply extrapolating from what a person did last time. The result, the company says, is a probability for each possible decision.
To build that picture, the system looks far beyond trades alone. It examines how a brokerage customer completes onboarding and identity checks, how often they call support, the tone of those calls, and how they transfer money and buy or sell. Those signals are then turned into mathematical representations of traits and fed into a transformer-based world model.
A separate language model writes explanations for the predictions, so a brokerage can see a reason alongside a forecast. During training, the model is rewarded for keeping a person’s tendencies stable in simulation, so a normally risk-averse customer should not suddenly be treated as a risk-taker. That consistency check is central to how 42AI describes its approach.
The numbers, carefully read
The headline figures need careful reading, and 42AI’s own founders supply the caveats. The company claims 98% accuracy in predicting market trades, but chief executive Amit Gal-Or clarified that the figure measures whether trading flows fall inside a predicted range, not whether every individual trade was called correctly. That distinction changes what the number means.
For predicting trading direction and volume, Gal-Or reported F1 scores of 0.8 to 0.85, a machine-learning measure that balances correct predictions against false alarms and missed events. One brokerage, he claimed, generated about $30 million in additional value through dynamic hedging, and the company says it is generating more than seven figures in revenue, without giving a precise number.
The funding also needs its structure attached. The $62.5 million was assembled over three years from the founders and Infinity Group, and was not entirely raised from external investors. Infinity Group was founded by Amir Gal-Or, the father of chief executive Amit Gal-Or, who is also 42AI’s chairman and a co-founder. The company now plans to raise more money.
Why It Matters
Brokerages live or die on anticipating flows, and a tool that forecasts client behavior could change how they set hedges, margin, and liquidity buffers. If the forecasts are reliable, a firm could react earlier to a rush for the exits or a buying wave, rather than discovering the move after it happened. That is the commercial promise here.
The same capability is also why this story will draw scrutiny. A brokerage that can anticipate what its customers will do holds an advantage over them, and 42AI’s founders did not explain a mechanism stopping clients gaining an edge. Regulations, monitoring, and an ethics committee were the safeguards offered, and they are process answers, not a structural fix.
The data scale sharpens that question. 42AI says its team first studied thousands of people through psychological assessments and behavioral scenarios, then added government and biobank datasets covering more than 2 million people. Its customer datasets have since expanded to cover almost 30 million people, according to Gal-Or, a figure that invites questions about consent and governance.
Our read
42AI’s strongest pitch is also its weakest: it claims to model why people act, not just what they did, yet its evidence is a range-coverage figure and an F1 range, not an independent audit. Until an outside party tests it on held-out customers and publishes the method, the 98% figure is a company claim, not proof that behavior is solved.
That caution does not make the launch trivial. Ten large brokerages, seven-figure revenue, and a three-year stealth build suggest a working product rather than a slide deck, and plans for gaming, fraud prevention, and cybersecurity show where it travels next. Each new domain needs its own training, as co-founder Ehud Ezra acknowledged, so brokerage results do not automatically transfer.
Watch the governance test next, not just the accuracy claims. If brokerages disclose how these profiles are built and what limits bind their use, the technology could become a risk-management tool. If they do not, it becomes an information advantage for the house over its customers, and regulators will ask who benefits from knowing what a customer will do next.
Background
42AI spent three years developing its approach, combining psychological research with neuroscience and genetics, and taking advice from psychology and post-traumatic stress specialists and from Dan Shechtman, who won the 2011 Nobel Prize in Chemistry. Its founders say the model has a natural ceiling, because it cannot see everything a person experiences.
Source: TechCrunch report on 42AI and Unite.AI coverage of the launch
0 Comments