Israeli startup 42AI emerged from stealth on November 2 with $62.5 million in funding and an ambitious proposition: an AI model intended to predict what people will do, rather than simply what they will write. According to TechCrunch, its initial application is inside financial brokerages, where it aims to estimate how individual traders will respond to market volatility.
The financing differs from the conventional seed-round story. CEO Amit Gal-Or told TechCrunch that the money came from the founders and Infinity Group over three years. It was not a single check from outside investors delivered on launch day. That distinction changes how the headline amount should be understood.
The technical promise also needs precision. The company describes probabilistic forecasts of behavior under particular conditions. It is not claiming certainty about every person in every situation. Its commercial challenge sits between those descriptions: turning an estimate of a possible response into information useful for a real decision.
What a behavioral world model means
A language model estimates which words or text tokens fit next in a sequence. 42AI wants to predict something different: a person’s actions in a changing environment. The question is not simply what a trader might say, but how likely that trader is to increase exposure, reduce it or wait as circumstances change.
The company calls its transformer-based system a behavioral world model. In its description, a separate language model can explain the behavioral predictions. The verbal explanation and the forecast are therefore distinct components: a persuasive explanation is not evidence that the next action was predicted well.
People also respond to information a system may never see. A financial need, a personal event or an unseen piece of information can change a decision. Two similar past actions may have different causes. A reasonable forecast therefore needs room for uncertainty, rather than presenting one path as inevitable.
Think of the difference between one answer and a range of possibilities. A system estimating several scenarios can help planning without identifying exactly which one will occur. Its value depends on whether the range is useful. A very wide forecast can include the eventual outcome while offering little help to someone who must act in time.
Why start with brokerages
At a brokerage, client responses can affect liquidity needs, margin requirements and hedging exposure. 42AI says estimating trader behavior can help manage those variables. This is an operational use of forecasts about customers, rather than a promise that customers will receive profitable recommendations for their next trade.
Consider a hypothetical sharp market move: one group reduces positions while another increases them. The brokerage might need different preparations for each scenario. The model would estimate possible responses; deciding what to do about them is a separate question. This illustration is not a commercial result demonstrated by the company.
That focus also narrows the problem. Rather than predicting all human behavior everywhere, the system operates within an activity whose events, actions and outcomes can be defined. Success there, if demonstrated, would not automatically establish success in gaming, security or human-robot interaction. The meaning of success changes across those domains.
The company is the source of the capabilities and uses it describes. That does not make the application irrelevant, but using a product is different from showing a consistent advantage over a simpler alternative. Deployment answers whether something can run; evaluation answers how well it performs.
The money accumulated, and the relationships explain its structure
Amit Gal-Or is CEO and co-founder, while Ehud Ezra is CTO and co-founder. Amir Gal-Or, Infinity’s founder and Amit’s father, is chairman and a co-founder. Those relationships help explain the reported financing structure and belong in an account of the funding.
TechCrunch describes the money as seed funding while explaining that it accumulated across three years of work. Both descriptions can coexist, provided readers are not left thinking a new outside fund invested the entire amount now. The headline tells them how much funding supported the operation, but not when it arrived or who supplied it.
The amount is not a measure of predictive accuracy either. Funding can support development, hiring and infrastructure; it cannot substitute for testing the forecasts. For technology intended to predict people, separating resources from evidence matters because the promise extends beyond performing a single clearly defined action.
What to ask before accepting an accuracy percentage
The report includes a company-supplied 98% figure. It concerns trading flows falling within a predicted range, not correctly predicting 98% of individual trades. Those are different measures. Without the range width, forecast horizon and baseline, the number alone does not establish how useful the prediction is.
Test conditions matter too. Results in calm markets do not necessarily establish results during an exceptional shock, and one set of traders may not represent every customer. A useful evaluation should explain which people and situations the result covers, including when the forecast stopped helping.
Privacy requires similar care. A company’s statement that it does not use personally identifiable information does not resolve every question about individual behavioral profiling. Business customers need to understand what data is used, for what purpose, who can access forecasts and which decisions depend on them. These are questions about use, not a claim that harm occurred.
42AI is also exploring gaming, fraud prevention and cybersecurity, and eventually wants to help robots and AI agents understand human behavior. Those remain stated directions. The nearer, more concrete test is inside brokerages: whether well-defined forecasts improve operational decisions under conditions where mistakes can also be measured.