On Tuesday, payment card provider Mastercard announced a new generative AI model trained on its payments data.

The “foundation” model, built in collaboration with Nvidia and AI data company Databricks, will serve as the basis for a “wide range of applications,” Mastercard says. It is a large tabular model, or LTM, meaning it has been trained on databases and large-scale tables. – unlike popular consumer models such as Claude or ChatGPT, which are trained on unstructured datasets including images, video, and text.

Mastercard says the model is being trained to predict future transactions by reviewing billions of anonymised previous ones. The company intends to use it to improve services across its offerings, including cybersecurity, small business tools and loyalty programmes.

The announcement details the model’s use in improving fraud detection: current AI models tend to flag large but infrequent purchases – such as wedding rings – as potential fraud, but the volume of data available to the LTM has made it less likely to do so in testing.

This is not Mastercard’s only foray into AI. It also offers a consumer-facing agentic AI purchasing tool, Agent Pay. Earlier this month, Mastercard partnered with Banco Santander to perform what it said was the first successful completion of a live end-to-end payment by an AI agent in Europe.

Mastercard also announced on Tuesday that it is acquiring London-based stablecoin infrastructure startup BVNK for up to $1.8 billion, its largest ever cryptocurrency related deal. This represents a significant premium on its $750 million valuation the company announced last year.

---