Israeli startup 42AI exits stealth with $62.5M for a behavioral world model of individual people

What happened

Israeli startup 42AI has emerged from stealth with $62.5 million in seed funding and one of the more unusual claims in current AI research: a behavioral world model that predicts how individual people will act, including in situations they have never faced before.

According to TechCrunch, the system is already live at ten large financial brokerages, where it forecasts how individual traders will respond to volatile conditions and feeds into decisions about liquidity, margin requirements and hedging exposure.

The funding has an unusual structure. TechCrunch reports the $62.5 million came from the founders and Infinity Group across three years rather than through a conventional external round. Infinity was founded by Amir Gal-Or — the CEO's father — who is also 42AI's chairman and a co-founder. The company says it intends to raise more.

Modeling tendencies rather than history

Co-founder and CEO Amit Gal-Or told TechCrunch that behavior is probabilistic, not deterministic, and that people carry psychological tendencies that keep expressing themselves even as circumstances change. The model's task is to detect those tendencies and output the probability of a given decision.

Rather than extrapolating from past actions, 42AI says it infers internal motivations from how people react to loosely related events — for a trader, everything from KYC onboarding and support calls to money transfers and trades. Co-founder and CTO Ehud Ezra said the system ingests sentiment, voice characteristics, call frequency and trading activity, then converts those observations into mathematical representations of traits.

Those trait representations feed a transformer-based world model that produces probability distributions over possible decisions in different scenarios, with a separate language model generating explanations for each prediction. Ezra said training rewards are designed to keep a person's baseline tendencies stable in simulation: if a normally cautious trader starts behaving like a risk-taker, the model is penalized.

Three years of research and data

TechCrunch reports 42AI spent three years building the approach, combining psychological research with neuroscience and genetics, with input from experts in psychology and PTSD as well as Dan Shechtman, who received the 2011 Nobel Prize in Chemistry.

The team began with psychometric assessments of thousands of people, then added government and biobank datasets covering more than 2 million people, according to Gal-Or. Customer datasets have since grown to cover nearly 30 million people.

In production, the model analyzes existing customer records such as transactions and support calls. Gal-Or argued it does not require personally identifiable information because the model never learns who a person is, though as TechCrunch points out, 42AI still builds individual behavioral profiles from linked customer activity.

Performance claims, with caveats

The company says its model reaches 98% accuracy when predicting market trades, but Gal-Or clarified that this refers to whether trading flows fall inside a predicted range, not whether each individual trade is called correctly. He said the model achieves an F1 score of 0.8 to 0.85 for predicting trading direction and volume.

The founders also acknowledged limits. Gal-Or described a natural ceiling since the system cannot observe everything a person experiences, and Ezra said accuracy drops when meaningful context is missing — for instance, a trader changing strategy on a friend's advice.

Commercial traction and open questions

According to TechCrunch, 42AI generates more than seven figures in revenue and charges either a flat fee or a share of the value it creates. Gal-Or claimed one brokerage produced roughly $30 million in additional value through dynamic hedging, a figure that is the company's own and not independently verified.

The company is exploring gaming, fraud prevention and cybersecurity as its next domains, and eventually wants robots and AI agents to use the model to understand people — though Ezra conceded that each new domain requires retraining on its specific rules and context.

There is also an obvious tension: if a brokerage can anticipate what its customers will do, the same knowledge could be turned against them. Ezra called that a valid concern, and the founders pointed to regulation, monitoring and an internal ethics committee, but as TechCrunch observes, they did not explain what would actually stop a customer from exploiting the profiles of the people being modeled. Gal-Or insisted the technology itself is part of the solution.

Why it matters

World models are one of the most active threads in AI research, but the work so far has focused overwhelmingly on physical environments, video and game simulation. 42AI is pointing similar architecture at individual humans — in Ezra's framing, each person is the world being modeled — which marks a notably different direction from the dominant language-model paradigm.

The story is also an early preview of the commercial and ethical friction this creates. If behavioral prediction works even partially, the first buyers are institutions with money at risk, not the people being profiled, and questions of consent, profiling and asymmetric information arrive long before any regulation does. The company's claims — particularly the 98% figure — warrant skepticism, and a funding round drawn largely from a family-linked investor means the wider market has not yet voted on the technology.