Mastercard's New AI Model Redefines Fraud Detection in Payments
Mastercard is building a new generative AI model trained on anonymised payment data. The system is designed to improve cybersecurity, fraud detection, and other financial services without relying on manual input. Early tests show it can outperform traditional machine learning in certain tasks.
The model is a large tabular system trained on structured transaction records. Unlike chatbot-style AI, it functions as an insights engine for payments and commerce. Mastercard has already processed billions of anonymised transactions to develop it and may expand training to other datasets over time.
Initial applications will focus on fraud prevention and cybersecurity. The company plans to combine this model with existing AI tools to create hybrid defence systems. Future uses could extend to loyalty programmes, personalised services, and advanced analytics.
Mastercard is working with Nvidia and Databricks on the project. More technical details will be revealed at Nvidia's GTC conference in 2026. The model's ability to learn patterns with less human intervention sets it apart from older systems.
Previous AI fraud detection efforts faced hurdles. In 2023, JPMorgan Chase's machine learning models struggled with false positives, initially hitting 10%. PayPal's 2022 graph neural networks encountered delays and adversarial attacks. These challenges provided Mastercard with benchmarks—aiming for false positives below 1% and stronger defences through generative AI.
The new model could change how financial institutions handle security and customer services. By reducing manual input and improving accuracy, it may set a higher standard for fraud detection. Mastercard's hybrid approach suggests broader applications in payments and analytics are likely.
