Mastercard has introduced a new type of artificial intelligence designed to improve security in digital payments. Instead of using traditional language or image-based models, they built a large tabular model that focuses solely on transaction data. This approach aims to catch fraud and verify authenticity while protecting user privacy.

What Is a Large Tabular Model?

A large tabular model, or LTM, works differently from typical language models. Instead of predicting words or sentences, it examines relationships between different pieces of structured data in tables. These tables include details like merchant location, transaction times, authorization steps, and chargebacks. Mastercard trained the model on billions of transactions to find patterns that could signal fraud or suspicious activity.

This model learns from raw data, identifying connections and anomalies without relying on personal identifiers. By focusing on behavioral patterns rather than individual identities, the technology reduces privacy risks often associated with AI in finance. Mastercard emphasizes that personal data is removed before training begins, making the system safer and more privacy-conscious.

How Mastercard Uses the Model

The company says the LTM acts as an “insights engine,” supporting existing fraud detection tools. It can analyze vast amounts of transaction data quickly and spot unusual patterns that might otherwise go unnoticed. For example, it can detect high-value purchases made infrequently or transactions that occur in distant locations within a short period.

Mastercard is currently deploying this technology in their cybersecurity systems. It helps improve the accuracy of fraud detection while reducing false alarms. Unlike some other AI systems that interact directly with customers, this model is mainly used internally to strengthen security protocols.

The infrastructure powering the LTM comes from Nvidia and Databricks. Nvidia provides the computing power needed to process large datasets, while Databricks handles data engineering and model development. This collaboration ensures the model runs efficiently and can scale as data volumes grow.

Potential Impact and Future Developments

Early results from Mastercard’s pilot programs show promising improvements over traditional methods. The model has demonstrated better performance in flagging suspicious transactions, especially in cases involving high-value, low-frequency purchases. This suggests it could help prevent financial fraud more effectively.

While anonymization removes some data signals that could be useful for risk assessment, Mastercard believes that the sheer volume of behavioral data compensates for this loss. They argue that larger datasets enable the model to learn complex patterns without needing personal information, which benefits privacy and security alike.

Looking ahead, Mastercard plans to expand the use of this technology beyond cybersecurity. The scalable nature of the LTM could also enhance fraud detection in other areas of digital payments and financial services, making transactions safer for everyone.

Inspired by

• https://www.artificialintelligence-news.com/news/mastercards-ltm-keeps-tabs-on-fraud-with-a-new-foundation-model/

Sources

• chatgpt.com