THE ESSENTIALS IN 30 SECONDS

  • 42AI's $62.5 million seed came from its founders and Infinity Group over three years, rather than from a conventional outside funding round.
  • The transformer-based model estimates how likely a person is to take a given decision; 10 large brokerages use it to manage liquidity, margin and hedging, the company says.
  • A claimed 98% accuracy measures whether trading flows land within a predicted range, not individual trades; none of the figures have been independently checked.

Israeli startup 42AI came out of stealth on November 2 with $62.5 million in seed funding and a product it calls a "behavioral world model": software that estimates how likely a specific person is to make a particular decision, including in situations the system has not seen before. Its first market is retail brokerage.

Funding and team

The round is unusual. As TechCrunch reports, chief executive Amit Gal-Or said the money came from the founders and from Infinity Group over three years, rather than from a single external round. Infinity was founded by Amir Gal-Or, 42AI's chairman and a co-founder; Ehud Ezra is chief technology officer. The company says it intends to raise more.

How the model works

42AI turns observed behaviour into mathematical representations of personal traits, which feed a transformer-based world model, while a separate language model writes explanations of each prediction. Early research involved thousands of people taking psychological assessments and behavioural scenarios; government and biobank datasets covering more than 2 million people were added later, and customer data now spans nearly 30 million people, according to the company. Inputs include onboarding and know-your-customer steps, the sentiment and frequency of support calls, money transfers and trades. 42AI says the deployed system does not need personally identifiable information, though TechCrunch points out it still builds a behavioural profile for each linked customer.

A preprint on arXiv from February, co-authored by Ezra, describes a "Large Behavioral Model" fine-tuned from Llama-3.1-8B-Instruct to predict individual strategic choices from psychometric profiles. It reports gains over the base model and results comparable to frontier models, but the abstract gives no headline accuracy numbers.

Claims to treat with caution

The company says it works with 10 large brokerages, which use its forecasts to manage liquidity, margin requirements and hedging, and that its revenue is in the seven figures. It cites one brokerage that gained about $30 million through dynamic hedging. It also claims 98% accuracy in predicting trades, but the CEO told TechCrunch this means trading flows fell within a predicted range, not that each trade was called correctly; F1 scores for direction and volume are put at 0.80 to 0.85. None of these figures has been independently verified.

The founders acknowledge limits: the model misses context such as advice from a friend. Asked whether brokerages could use predictions against their own clients, Ezra called it a valid concern, pointing to regulation, monitoring and an internal ethics committee without describing specific safeguards. 42AI is also exploring gaming, fraud prevention and cybersecurity.