Pomegra
Tel Aviv-based 42AI came out of stealth on November 2, 2026 with $62.5M in founder and Infinity Group capital for a model that predicts human behaviour.
Key Takeaways
- 42AI disclosed $62.5M in seed funding, raised from its founders and Infinity Group over three years.
- No outside venture fund led the round, and no valuation was disclosed.
- The model runs at 10 brokerages and covers about 30 million people, with seven-figure revenue.
Lead
42AI launched publicly on November 2, 2026, with a behavioral world model that estimates the probability of each choice a person might make in a given situation. The company reports $62.5M in seed funding, accumulated over three years. The money came from the founders and Infinity Group, a firm founded by Amir Gal-Or, who is 42AI's co-founder and chairman. His son, Amit Gal-Or, is co-founder and CEO, and Ehud Ezra is chief technology officer.
The startup is headquartered in Tel Aviv. Its advisers include Dan Shechtman, the 2011 Nobel laureate in chemistry.
What Does 42AI's Model Actually Do?
42AI's system converts psychometric profiles into mathematical representations of personality traits, then feeds them into a transformer-based world model. A separate language model explains what drove each prediction. The company has said it adapted Meta's Llama 3.1, an 8-billion-parameter model, for the work.
The core claim is that behaviour is probabilistic rather than fixed. People carry stable psychological tendencies that persist when circumstances change, the company argues, so the model tries to infer motivation instead of extrapolating from past actions. Training drew on datasets covering more than 2 million people during development. Customer data now spans about 30 million individuals.
Who Is Paying for It?
Ten large brokerages use the model today, according to the company. They apply it to liquidity management, margin requirements and hedging exposure, mostly by forecasting how clients trade when markets turn volatile. Inputs include know-your-customer checks, client call frequency, voice sentiment and linked account activity.
Revenue is seven figures, with the exact amount undisclosed. Pricing is either a flat fee or a share of the value generated. 42AI says one brokerage gained about $30M through dynamic hedging.
How Accurate Is the 98% Claim?
The 98% figure measures whether aggregate trading flows land inside a predicted range. It does not measure whether the model calls each individual trade correctly. That is a weaker test than the headline suggests, and 42AI has said so itself.
The company also reports F1 scores of 0.8 to 0.85 for predicting trade direction and volume. Those numbers have not been independently verified. Outside researchers have noted that the privacy claim, "the model never knows who the person is," should be tested for re-identification risk, since individual profiles are still built from linked activity.
What Does This Round Say About 42AI's Funding?
A $62.5M seed with no outside venture lead is unusual. Most rounds of that size carry a named institutional investor and a priced valuation. Here, the capital is insider money spread across three years, which means no fund has yet set a price on the company.
That cuts two ways. The founders kept control and avoided early dilution, but the model has not faced diligence from a growth investor. 42AI has said it plans to raise more money, and that raise will be the first external test of its claims.
Where Could the Technology Go Next?
42AI plans to extend the model beyond brokerages into gaming, fraud prevention and cybersecurity. Its longer-term aim is to help robots and AI agents understand people. Each of those markets has incumbents with entrenched scoring and detection systems, and each will demand evidence beyond range-level accuracy.
Regulatory exposure follows the data. Behavioral prediction applied to advertising, hiring, insurance or political targeting would draw scrutiny under privacy and automated-decision rules in the EU and several US states. Brokerage use is narrower, but it still involves profiling clients.
Outlook
42AI has real deployments, seven-figure revenue and a self-funded war chest, which is more than most stealth-stage AI startups can show. What it lacks is an external valuation, independent accuracy testing and a customer base outside finance. The next funding round and any third-party audit of the 98% claim will show whether the model holds up beyond its first ten clients.
