Mastercard is building a new generative AI model using anonymised transaction data, with cybersecurity and fraud detection among its first planned uses.

The company said the model is designed as an insights engine for payments and commerce rather than a chatbot. It described the system as a large tabular model trained on structured data.

Over time, the model could also support services such as loyalty and rewards programmes, personalisation, portfolio optimisation and analytics.

Mastercard said it is training the system on billions of anonymised transactions and may later expand it to include other datasets such as merchant location, fraud, authorisation, chargeback and loyalty programme data.

The company said the model can learn patterns from data with less manual input than its existing systems, which often rely on data scientists to add features that help flag suspicious activity.

In early testing, Mastercard said the model outperformed standard machine learning techniques in some cases, including showing better ability to identify legitimate high-value purchases that might otherwise be flagged as suspicious.

Steve Flinter, Distinguished Engineer at Mastercard, said,

“We plan to build hybrid cybersecurity systems that combine the best of both our current AI models and this new LTM.

This should help us build up and futureproof our cyber defenses.”

Mastercard said the model could eventually help reduce the need to build and maintain thousands of separate AI models across different markets, customers and use cases.

The project is being developed with Nvidia and Databricks, and Mastercard plans to share more about the work at Nvidia GTC 2026.

Featured image: Edited by Fintech News Singapore, based on image by Frolopiaton Palm via Freepik