42AI Exits Stealth With $62.5M to Predict What You'll Do Before You Decide

An Israeli startup called 42AI emerged from stealth mode this Monday (November 2) with a $62.5 million funding round and technology that sounds more like science fiction than enterprise reality: a "behavioral world model" capable of predicting how a specific person will react to novel situations, even before they themselves decide what to do.

The Tel Aviv-based company isn't building the next chatbot or text generation tool. Its proposition is radically different — and potentially more controversial. Instead of modeling what people say they want, 42AI wants to model what they actually do. The gap between declared intention and real action is exactly where the company believes the value lies.

What 42AI is actually building

Most artificial intelligence that people use today runs on large language models. These systems learn from enormous quantities of text and operate based on what you ask for. 42AI is following a different path entirely.

According to CEO and co-founder Amit Gal-Or, the company's model is not deterministic — it's "constantly probabilistic." The idea is that people have underlying psychological tendencies that influence their decisions even when circumstances change, and that's what the startup is trying to measure and translate into data a model can understand.

"A week from now, you might be a different person, but your tendencies will always manifest in certain areas as a probability or distribution," co-founder and CTO Ehud Ezra added.

The model works by converting observations about each individual into mathematical representations of traits, which are then fed into a transformer-based world model to generate probability scores for possible decisions under different circumstances. A separate language model is used to generate explanations for these predictions.

How the data is collected

42AI says it trained its model by analyzing how people react to seemingly unrelated events. For an investor or trader, these events include completing KYC (Know Your Customer) processes, onboarding, support calls, money transfers, and stock trades.

"When the person is talking to support or sales during the initial process, we look at sentiment, voice, frequency of calls, and so on," Ezra said. All of these data points go into the model. Later, the model observes how and when people buy and sell, integrating real-time market data.

The company says its team initially studied thousands of people through psychological assessments and behavioral scenarios before adding government and biobank datasets covering more than 2 million people. The expanded customer datasets now cover nearly 30 million people.

42AI insists the system does not need personally identifiable information, arguing that "the model never knows who the person is." In practice, however, the company creates individual behavioral profiles from linked customer activity.

Where the technology is already deployed

42AI's initial application is in financial markets. The company says its model is already running at 10 major brokerages, helping manage liquidity, margin requirements, and hedging exposures. One of the clients, according to Gal-Or, generated approximately $30 million in additional value through dynamic hedging.

Clients pay the startup a flat fee or a percentage of the value the platform helps capture. The founders told TechCrunch the company is generating more than "seven figures" in revenue, though they declined to specify an exact figure.

In trading, 42AI claims its model can predict individual traders' actions — instrument, direction, timing, and volume — at horizons of one, four, and 24 hours, with accuracy above 98% across millions of traders. The company clarified, however, that this figure is more a measure of whether trading flows fall within a predicted range, not whether each individual trade was predicted correctly. The model consistently achieves an F1 score of 0.8 to 0.85 in predicting trading direction and volume.

What separates 42AI from the rest of the hype

What makes 42AI's case different from the rest of the world models hype is its funding: the $62.5 million was not raised in a conventional VC round. It came from the founders and the Infinity Group over three years. The Infinity Group was founded by Amir Gal-Or, Amit's father, who is also 42AI's co-founder and chairman. The structure resembles more a founder-backed vehicle than a traditional seed round with a term sheet from a venture fund.

Additionally, the technical approach is distinct from the LLMs dominating the market. While language models depend on declared intentions, 42AI builds around real behavior. A February 2026 research paper (arXiv:2602.17222), titled "Decoding the Human Factor: High Fidelity Behavioral Prediction for Strategic Foresight," describes a Large Behavioral Model (LBM) built through fine-tuning of Llama 3.1-8B-Instruct with structured psychometric profiles. The paper shows the adapted model outperforms its unmodified backbone in held-out scenarios, and that richer trait profiles offer benefits beyond prompting with broad personality categories.

The team is small — between 11 and 50 employees — and the company is registered in Delaware, a common configuration for Israeli startups seeking American capital.

Risks and limits

42AI's technology raises serious privacy concerns that warrant scrutiny. If a brokerage can anticipate what its clients will do, what prevents it from using that knowledge to benefit itself at the expense of the very people being profiled?

The founders admitted it's "a valid concern," pointing to regulations, monitoring, and an internal ethics committee, but did not explain what would actually stop clients from using the technology to gain advantage over the people being mapped.

And there's a natural ceiling to accuracy. "You really don't have access to everything a person experiences," Gal-Or said. The model's accuracy is also affected when it misses important situational context — such as when someone changes their trading choices based on advice from a friend or colleague.

Where 42AI wants to go

Beyond finance, 42AI is exploring applications for its model in gaming, fraud prevention, and cybersecurity. Ultimately, the company wants robots and AI agents to use its model to understand people.

The question is whether the idea of "predicting human behavior" will become a legitimate enterprise tool or a Pandora's box of manipulation and surveillance. 42AI has not disclosed information about public product launches or announced customer partnerships. For now, the applications of its technology remain largely invisible to the market.

42AI's behavioral world model is an ambitious experiment still without external benchmarks. Human behavior is, at best, complex — and probabilistic, as the founders themselves acknowledge. What the company has not shared is equally notable: no information about public product launches or announced customer partnerships. The market still has no way to independently assess whether the model works as well outside of trading rooms as it appears inside them.

The question remains: when a machine can predict your actions before you decide them yourself, who benefits — the brokerage or you?

Sources: TechCrunch, CryptoBriefing, Unite.AI

✓ Independent sources cross-checked and verified before publishing