42AI exits stealth with $62.5M for a 'behavioral world model' that predicts what traders will do

Israeli startup 42AI came out of stealth today with $62.5 million in seed funding and a claim straight out of science fiction: a “behavioral world model” that can estimate what a person will do, even in a situation that person has never faced. The company spent three years building the system in stealth, and its first customers are financial brokerages, where the model tries to predict how individual traders will respond when markets turn volatile.

The funding story needs the same care as the technology claim. The $62.5 million came from the founders and Infinity Group over three years and was not entirely raised from external investors, chief executive and co-founder Amit Gal-Or told TechCrunch. Infinity was founded by his father, Amir Gal-Or, who is also 42AI’s chairman and a co-founder. The company now plans to raise more money from outside investors.

Probabilities, not certainties#

“Behavior is not deterministic; it’s constantly probabilistic,” Gal-Or told TechCrunch. His argument is that people carry underlying psychological tendencies that keep influencing their decisions even as circumstances change, and that those tendencies can be measured and turned into data a model can use. “A week from now, you might be a different person, but your tendencies will always play out in certain areas to a probability or distribution,” co-founder and chief technology officer Ehud Ezra added. What the model outputs, the company says, is the probability of a person making a certain decision.

42AI says its model tries to work out people’s internal motivations rather than extrapolating from their past actions alone. Observations are converted into mathematical representations of traits, which feed a transformer-based world model that lays out probabilities for possible decisions under different circumstances. A separate language model generates explanations for the predictions. Training is structured to preserve a person’s underlying tendencies during simulations: if someone is usually risk-averse and the model suddenly treats them as a risk-taker, Ezra said, “the target is off.”

What the model watches#

The inputs are deliberately drawn from events that look unrelated to trading. For an investor or trader, 42AI examines how the person reacted to completing know-your-customer checks, onboarding, support and sales calls, money transfers and trades. “When the person is talking to support or sales in the initial process […] we’re looking at the sentiment, we’re looking at the voice, we’re looking at a lot of interactions that they’re doing, how many times do they call, and so forth,” Ezra said. “And all of those data points are going into the model.”

The approach took three years to develop, combining psychological research with neuroscience and genetics, plus advice from psychology and PTSD experts and Dan Shechtman, who won the 2011 Nobel Prize in Chemistry. The 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, Gal-Or said. In live deployments the model analyses customers’ existing records, including transactions and support calls. Gal-Or argued the system does not need personally identifiable information — “the model never knows who the person is” — although, as TechCrunch noted, the company still creates individual behavioral profiles from linked customer activity.

The numbers — and what they actually measure#

42AI claims 98% accuracy in predicting market trades, but Gal-Or clarified to TechCrunch that the figure measures whether trading flows fall within a predicted range, not whether each trade was predicted correctly. He also said the model consistently achieves an F1 score of 0.8 to 0.85 for predicting trading direction and volume — a machine-learning metric that balances correct predictions against false alarms and missed events.

On the commercial side, Gal-Or said 42AI is working with 10 large brokerages, which use the system to manage liquidity, margin requirements and hedging exposure, and is onboarding more customers. Revenue is more than “seven figures,” he said, without giving a precise number. Customers pay a flat fee or a percentage of the value the platform helps them gain; he claimed one brokerage generated approximately $30 million in additional value through dynamic hedging. Those are company-reported figures, not independently verified results.

Gal-Or was candid about the ceiling. “There’s a natural ceiling because you really don’t have access to everything that a person experiences,” he said. Accuracy also suffers when the model misses important situational context — Ezra’s example was someone changing a trading choice on a friend’s or colleague’s advice, something the system cannot see.

Beyond the trading floor — and the uncomfortable question#

Beyond finance, 42AI is exploring gaming, fraud prevention and cybersecurity, and one day wants its model to help robots and AI agents understand people. That is, by the founders’ own account, some way off: Ezra acknowledged the model must be trained on the rules and context of each new domain before it can work there.

Then there is the question TechCrunch put to the founders directly: if a brokerage can anticipate what its customers may do, what stops someone using that knowledge to their own benefit? Ezra admitted it was “a valid concern.” The founders pointed to regulations, monitoring and an internal ethics committee, but did not explain what, if anything, would stop their customers using the technology to gain an advantage over the very people it profiles. Gal-Or was confident the technology could help address such risks: “We are very proud that actually we are part of the solution; we are the solution to this problem.”

What 42AI has announced today is a funded exit from stealth with live brokerage deployments and striking, company-reported numbers. What it has not published is independent validation of its predictions, the terms on which brokerages use the profiles, or results outside finance. Until those arrive, the 98% figure should be read exactly as Gal-Or defined it: flows landing inside a predicted range, from a model that knows a great deal about how people behaved — and nothing about what they will hear from a friend tomorrow. Unite.AI’s summary of the announcement corroborates the funding, brokerage count and performance figures reported by TechCrunch.