The Tel Aviv startup is betting that models trained on real actions, not just text, can anticipate decisions people haven't made yet
An Israeli startup wants to know what you’ll do before you do. Specifically, it wants to know even when you’re facing a situation you’ve never encountered.
42AI, based in Tel Aviv, has come out of stealth with total funding of approximately $62.5 million to $63 million. Its pitch centers on something it calls Large Behavioral Models, or LBMs. The goal is to predict how people and institutions will act.
What 42AI is actually building
Most of the AI people use today runs on large language models. These systems learn from enormous piles of text. When you ask one for help, it works from what you tell it you want.
42AI is taking a different route. Its models are trained on real choices and real actions, not only on words.
Language models largely depend on stated intentions. 42AI is building around actual behavior. The company wants to model what people do, rather than what they claim they’ll do.
CEO and co-founder Amit Gal-Or framed the core idea in an interview with TechCrunch.
“Behavior is not deterministic; it’s constantly probabilistic,” Gal-Or said.
Put more simply, 42AI isn’t promising a crystal ball. It’s promising odds. The model doesn’t claim to know exactly what you’ll choose, only how likely each option is.
The company’s scope also extends beyond individuals. According to the research findings, 42AI aims to predict institutional actions as well as human ones.
From seed round to stealth exit
Before this announcement, 42AI completed a $5 million seed round. Its total funding now sits in the range of approximately $62.5 million to $63 million.
The team remains small. The company lists between 11 and 50 employees. It is based in Tel Aviv and registered in Delaware, a common setup for Israeli startups courting US capital.
On LinkedIn, 42AI describes its mission in terms of tackling what it calls the “Alignment Trap” in agentic systems. Agentic systems are AI tools that take actions on a user’s behalf, not just answer questions. Think booking flights or managing workflows rather than drafting an email.
The alignment problem in that context is straightforward to describe and hard to solve. An agent acting for you needs to do what you actually want. What you say you want and what you actually want are not always the same thing.
What the company has not shared is equally notable. 42AI has not released information about public product launches or customer partnerships. For now, the applications of its technology remain largely out of view.
What this means
The most immediate signal is about where investor money is flowing. A startup of this size raising tens of millions around a single concept suggests backers see behavioral modeling as a distinct category. It isn’t being pitched as a feature bolted onto a chatbot.
The competitive question is data. Training on real choices and actions requires access to large volumes of behavioral information. Where that data comes from, and under what terms, will likely shape how far a company like this can go. 42AI hasn’t detailed its data sources, so that remains an open question.
The absence of announced customers means the market has no external benchmark yet for how well LBMs perform.
Gal-Or’s own framing offers a fitting lens for judging the company. Behavior is probabilistic, he argues. So, for now, is 42AI’s success.
