Mastercard Uses AI Foundation Model to Detect and Prevent Credit Card Fraud

Mastercard has developed a large tabular model (LTM) trained on transaction data rather than text or images to address security and authenticity challenges in digital payments. Unlike traditional large language models (LLMs), this innovative approach focuses on structured financial data to enhance fraud detection and payment verification.

The company has trained this foundation model on billions of card transactions, with plans to expand to hundreds of billions over time. The datasets encompass payment events and associated information including merchant locations, authorization flows, fraud incidents, chargebacks, and loyalty activity. Importantly, personal identifiers are removed before training begins, ensuring the model analyzes behavioral patterns without accessing individual identities.

By excluding personal data, the technology significantly reduces privacy risks commonly associated with AI applications in the financial services sector. The scale and richness of the anonymized data enable the model to identify commercially valuable patterns while maintaining user privacy. Although anonymization removes certain signals that could be useful for risk assessment, Mastercard asserts that leveraging sufficiently large volumes of behavioral data compensates for this limitation.

Understanding Large Tabular Models (LTMs)

LTM architecture differs fundamentally from large language models. While LLMs process unstructured inputs and predict the next token in a sequence, Mastercard's LTM examines relationships between fields in multi-dimensional data tables. This approach aligns more closely with traditional machine learning rather than generative artificial intelligence.

The large tabular model learns directly from raw inputs to identify predictable relationships, enabling it to detect anomalous patterns that predefined rules might miss. Mastercard describes the LTM as an 'insights engine' that integrates with existing products and augments current workflows. The operational risk profile differs from customer-facing LLMs, as this model primarily supports internal decision-making processes.

The technical infrastructure leverages Nvidia for computing platforms and Databricks for data engineering and model development, combining industry-leading technologies to power the LTM.

Practical Applications and Deployment

Cybersecurity represents the first active deployment area for this technology at Mastercard. The company operates multiple fraud detection systems that examine transaction data, traditionally requiring human input to define suspicious behavior patterns such as sudden transaction frequency increases or purchases across different geographic locations within short timeframes.

Early results demonstrate improved performance over conventional techniques in specific scenarios. For example, the model shows enhanced accuracy in distinguishing legitimate high-value, low-frequency purchases from fraudulent ones—transactions that traditional models often flag as anomalies.

Mastercard plans to implement hybrid systems combining established procedures with the new model, reflecting the cautious approach required by regulatory frameworks. The company acknowledges that no single model excels in all scenarios, positioning the LTM as one tool among many in their security arsenal.

Beyond fraud detection, the model can analyze loyalty program activity, support portfolio management, and enhance internal analytics—any area involving large volumes of structured data. Currently, companies often deploy multiple specialized models for each task, multiplying training costs and validation efforts. A single foundation model that can be fine-tuned for different applications may streamline operations and reduce expenses.

Risks and Future Development

The multi-function LTM approach carries inherent risks: a failure in a widely-deployed model could have system-wide consequences. This explains Mastercard's strategy of deploying the technology alongside existing detection systems, at least initially.

Future plans include increasing the scale of training data and overall model sophistication. Mastercard is also developing API access and SDKs to enable internal teams to build new applications on the platform.

The company emphasizes its commitment to data responsibilities, including privacy protection, transparency, model explainability, and auditability. Regulatory scrutiny of systems influencing credit decisions or fraud outcomes is expected, alongside oversight of data practices involved in the LTM's operation.

The Future of Tabular AI in Financial Services

Highly structured data forms the foundation of the LTM approach. Large tabular models may represent the beginning of a new generation of AI systems in core banking and payments infrastructure. However, evidence to date remains limited to vendor reports, so performance claims should be evaluated carefully.

Critical factors will determine the success of tabular models: robustness under adversarial conditions, long-term post-training costs, and regulatory acceptance. These elements will shape the pace and extent of adoption across the financial services industry. For now, Mastercard is placing strategic bets on this emerging technology as part of its innovation roadmap.

Image source: "Oversight" by United States Marine Corps Official Page is licensed under CC BY-NC 2.0.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and co-located with other leading technology events. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.