Mastercard is developing a new artificial intelligence foundation model trained on billions of anonymized transactions, aiming to enhance payments, cybersecurity and customer experiences across its network.

The company said the model is designed as an “insights engine” rather than a chatbot, using structured financial data to predict transaction patterns and improve decision-making across services such as fraud detection, loyalty programs and analytics.

To build the system, Mastercard is using a type of deep learning architecture known as a large tabular model (LTM), which differs from the large language models used in popular chatbots. The model is being trained on transaction data stripped of personal identifiers, with plans to expand to hundreds of billions of records, including fraud signals, merchant data and authorization activity.

The effort is supported by AI infrastructure from Nvidia and data capabilities from Databricks, enabling large-scale processing and analysis.

Mastercard said early testing shows the model can outperform traditional machine learning approaches in certain scenarios, including reducing false positives in fraud detection. For example, the system demonstrated improved accuracy in distinguishing legitimate high-value purchases from suspicious activity.

The company plans to integrate the model into hybrid cybersecurity systems that combine existing AI tools with the new approach to strengthen fraud prevention and long-term resilience.

Beyond security, Mastercard expects the technology to support more personalized retail experiences, optimize financial portfolios and streamline analytics. The company also sees potential to reduce the need for maintaining thousands of separate AI models by replacing them with a more flexible, unified system.