42AI has emerged from stealth with $62.5 million in seed funding and a model intended to estimate how people may act under changing circumstances. In TechCrunch's report, the founders describe an initial focus on financial brokerages rather than a generally available system that can predict any person's decisions.
What the system models
The company describes a transformer-based world model that receives mathematical representations of behavioral tendencies. Its inputs can include transactions, onboarding activity and support interactions. A separate language model produces explanations for the predictions.
The founders frame the output as probabilities of decisions, rather than deterministic knowledge of what someone will do. They also acknowledge that missing situational context can reduce accuracy. The report notes that linked customer activity still creates individual behavioral profiles even if the model does not receive direct identifying information.
How to read the reported results
42AI claims 98% accuracy for trading flows falling within a predicted range. That is different from correctly forecasting 98% of individual trades. The company also reports an F1 score between 0.8 and 0.85 for trading direction and volume; the captured report does not provide an independent reproduction of those figures.
The seed funding accumulated over three years from the founders and Infinity Group. TechCrunch reports that it was not entirely raised from outside investors. The company says it works with ten large brokerages on liquidity, margin and hedging questions.
The questions beyond prediction quality
A practical assessment should ask how predictions are calibrated, what information they require and whose interests a deployment serves. The founders identify gaming, fraud prevention and cybersecurity as possible future directions, but say a new domain requires its own rules and context. The finance deployments therefore do not establish broad transfer to robots, agents or other industries.