Mastercard keeps tabs on fraud with new foundation model

Mastercard has developed a large tabular model (LTM) trained on billions of anonymized transaction records to enhance security and fraud detection in digital payments, marking a shift from traditional text or image-based AI models. Unlike large language models (LLMs), this LTM focuses on structured behavioral data such as merchant location, authorization flows, and chargebacks, enabling it to identify patterns indicative of fraudulent activity without relying on personal identifiers, thereby reducing privacy risks. This innovative approach leverages the scale and richness of transactional data to infer valuable behavioral patterns, compensating for the absence of individual-specific information. Mastercard