Mastercard has unveiled a new generative AI foundation model trained on billions of anonymized transactions, which the company says it plans to use as an “insights engine” for payments and commerce.
The company clarified the model is a backend infrastructure tool, rather than a consumer-facing chatbot.
It is being trained on transaction data to support a wider set of tools, from cyber defenses to loyalty programs and small-business services. Mastercard said the work is being developed with capabilities from Nvidia and Databricks.
From fraud tools to a foundation model
Mastercard has spent the past two years publicly tying generative AI to fraud and cyber defense. In February 2024, the company said Decision Intelligence Pro would scan one trillion data points and improve fraud detection rates by an average of 20%, and by as much as 300% in some cases, while reducing false positives by more than 85% in Mastercard’s own analysis.
In May 2024, Mastercard said a separate generative AI-based predictive technology doubled compromised-card detection rates and increased the speed of identifying at-risk merchants by 300%. Those products were narrower, risk-specific applications. The March 17, 2026 post describes a foundation model Mastercard expects to use across multiple product lines.
Where the foundation model fits Mastercard’s revenue mix
The shift toward foundation models mirrors Mastercard’s growing reliance on its Value-Added Services (VAS) segment.
In its 2025 Form 10-K, the company said value-added services and solutions generated $13.315 billion in net revenue, up from $10.832 billion in 2024, and said those services primarily include security solutions, consumer acquisition and engagement services, business and market insights, digital and authentication solutions, processing and gateway, ACH batch and real-time account-based payments and solutions, and open finance.
That description aligns more closely with Mastercard’s value-added services business than with the core card network alone.
What Visa and PayPal are building in the same space
Mastercard is also not making this move in a vacuum. Visa has been widening its own AI push across both fraud and AI commerce. In March 2024, Visa said its growing services business was adding three new AI-powered risk and fraud prevention products under Visa Protect.
In April 2025, Visa announced Visa Intelligent Commerce, saying it would open its payments network to developers and engineers building AI commerce and collaborate with companies including Anthropic, IBM, Microsoft, Mistral AI, OpenAI, Perplexity, Stripe and Samsung. In October 2025, it followed that with Trusted Agent Protocol, which Visa and Cloudflare said was designed to help merchants recognize trusted agents and verify their credentials.
PayPal has taken a related route. In October 2025, it launched agentic commerce services spanning an agentic payment solution plus catalog and order-management tools, then in January 2026 said it was powering Microsoft’s launch of Copilot Checkout.
Mastercard’s own agentic commerce moves and what remains undetailed
Mastercard itself has already moved in that direction through Agent Pay in April 2025 and Agent Suite in March 2026, both tied to agentic commerce and AI-driven operations. Mastercard has not yet detailed how the new foundation model will be deployed across those efforts.
The broader promise in the March 17 announcement is wider: a model trained on Mastercard’s transaction data that the company says could improve multiple services.