42AI lands $62.5M to predict human decisions with a behavioral world model

What if an AI could calculate what you were about to do before you even made up your mind?

That sounds like pure speculative fiction, but Israeli startup 42AI is turning it into a pitch deck. On November 2, the company exited stealth with a $62.5 million seed round, backed over three years by its founders and Infinity Group. Its flagship product is what it calls a "behavioral world model," built specifically to anticipate human choices under pressure.

Decoding psychology from support logs

Traditional financial models track your trading history and extrapolate the trend line. 42AI argues that approach misses the point. According to co-founder and CEO Amit Gal-Or, human decisions are probabilistic rather than deterministic, driven by underlying psychological traits that stay relatively stable even when circumstances shift.

To map those traits, 42AI looks at seemingly disconnected touchpoints. The system monitors how a user navigates onboarding and KYC checks, analyzes vocal sentiment and call frequency during customer support chats, and tracks transfer habits before tying them back to live trading actions.

CTO Ehud Ezra puts it simply: while a robotics world model treats the physical environment as the world, here "every individual is the world." These interaction signals are translated into mathematical vectors and fed into a transformer-based world model. The system simulates decision distributions across various market shocks, while a separate language model drafts natural-language explanations for each forecast.

The startup spent three years crafting the engine alongside psychology and PTSD researchers, plus 2011 Nobel chemistry laureate Dan Shechtman. Starting with psychological assessments on thousands of test subjects, the team folded in government and biobank records covering over 2 million individuals, eventually scaling to client datasets covering roughly 30 million people.

98% accuracy, with several asterisks

The technology is already live at 10 large financial brokerages, helping them optimize liquidity, dynamic hedging, and margin requirements. Gal-Or says the business is pulling in seven-figure revenues through flat fees or performance cuts, claiming one brokerage unlocked $30 million in added value via dynamic hedging.

Yet the headline numbers require a closer look. 42AI touts a 98% accuracy rate, but Gal-Or clarified that this only measures whether broader trading flows fall within an expected range, not whether the model calls every single execution. For predicting trade direction and volume, the system posts an F1 score between 0.8 and 0.85.

There is also a hard ceiling. If an investor flips their strategy based on a watercooler tip from a colleague, the model cannot capture that external context, and its accuracy drops accordingly.

Then comes the elephant in the room: front-running. If a brokerage can reliably model client panic or risk appetite minutes ahead of time, what stops it from trading against them? The founders pointed vaguely to regulatory scrutiny and an internal ethics committee, but offered no concrete technical safeguards preventing predatory use.

Why it matters

We see 42AI's pivot as one of the shrewdest commercial bets on world models to date.

While frontier labs spend billions teaching models Newtonian physics so robots can fold laundry, 42AI realized that human market sentiment is an equally potent dynamical system. In volatile markets where retail traders act on fear and FOMO, anticipating liquidity crunches gives brokerages an enormous operational edge.

For everyday traders, however, this should set off alarm bells. You are no longer just fighting faster fiber cables and latency arbitrage; you are trading against an institution that calculates your breaking point before you hit submit. Anonymized data offers little comfort when the system models your behavioral archetype down to your voice tone on support calls.

The real test will be generalization. 42AI is already eyeing gaming, fraud detection, and AI agents. But financial markets provide clear, structured feedback loops. Whether psychological vectors can reliably steer robots or customer agents across messier everyday domains remains an unproven gamble.