Fraud Detection · Risk Management
Mastercard wants to turn its transaction data into a reusable AI insights engine for payments and commerce, especially to strengthen cybersecurity and fraud detection, improve personalization and loyalty use cases, and reduce the operational burden of building, training, and maintaining thousands of separate market-, customer-, and use-case-specific AI models.
Reported impact
Mastercard reports strong early testing results, including performance above standard industry machine learning techniques, with the model better identifying legitimate high-value but infrequent purchases and the potential to strengthen cyber defenses while eventually reducing the need to maintain thousands of separate AI models.
Mastercard is building an internal foundation model for payments and commerce around a large tabular model (LTM) trained on structured, anonymized transaction data rather than the unstructured data used by chat-oriented LLMs. The implementation starts with billions of transactions, with planned expansion to hundreds of billions of payment events plus merchant location, fraud, authorization, chargeback, and loyalty program datasets. Personal data is removed before training. Nvidia's advanced accelerated computing platform provides the compute backbone that processes these large datasets at high speed, while Databricks is part of Mastercard's development stack and collaboration model. The LTM learns patterns from raw and lightly guided tabular data, then feeds insights into hybrid cybersecurity systems and other Mastercard tools. Mastercard is also building APIs and toolkits so internal teams can consume the model as a shared insights service across cyber, loyalty, analytics, and commerce applications.
Mastercard removes personal data from transaction records, then trains its large tabular model on billions of anonymized transactions and plans to add merchant location, fraud, authorization, chargeback, and loyalty datasets. Nvidia-powered compute processes the data for model training, and the resulting model is exposed through APIs and toolkits so Mastercard teams can embed its predictions and insights into cybersecurity tools and other commerce applications.
"Additionally, to run our network, we currently need to build, train and maintain thousands of AI models, each for different markets, use cases or customers."
"In our testing, we’ve already seen this new model outperform standard industry machine learning techniques, giving us promising early signs."
"This work is powered by Nvidia’s advanced accelerated computing platform."
"To protect user privacy, we remove all personal data from these transactions."
"We are training our foundation model on Mastercard’s unique datasets, starting with billions of transactions."
"We plan to use this new foundation model — not to build a chatbot — but as an insights engine that will make many of our tools and services even better, from cyber defenses to loyalty programs to small business tools."
"We will also continue our close collaboration with Nvidia and Databricks to push forward all this development work."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Jun 2, 2026 Revolut Revolut needed to replace siloed fraud, credit, marketing, and recommendation models that depended on months of feature engineering with a single behavioral foundation model that could learn from raw financial event… 16.1% improvement in credit risk prediction performance from NVIDIA Llama-Nemotron-Embed-1B-v2 NVIDIA H100 GPUs · NVIDIA Llama-Nemotron-Embed-1B-v2 · Nebius AI Cloud 
In-house Sample Sep 24, 2026 Mastercard Fraudsters are using AI to find vulnerabilities, adapt tactics, and scale attacks across more payment channels, while traditional transaction-by-transaction controls struggle to catch coordinated fraud and authorized… Last year, scam losses globally were estimated at $442 billion 
IBM Sample Feb 13, 2026 BORICA AD BORICA needed to modernize national fraud prevention so it could stop fast, cross-channel attacks in milliseconds without disrupting legitimate payment traffic across multiple banks and service providers. 4,000+ transactions per second IBM Safer Payments 