Meet Mastercard’s new generative AI model

Generative AI completely changed the world of chatbots. Now, talking to a robot isn’t sci-fi, it’s just something we do.

We believe this same gen-AI technology won’t just transform chat — it will transform commerce. It will make payments faster, retail experiences more personalized and cybersecurity tools more precise.

To make this vision happen, our teams have been researching and building a new foundation model, which is a large-scale AI model that can be used as a basis for a wide range of applications. We are training our foundation model on Mastercard’s unique datasets, starting with billions of transactions.

To protect user privacy, we remove all personal data from these transactions. After analyzing enough of these anonymized transactions, this model can start to predict future transactions.

This kind of prediction model is largely the same as how today’s chatbots work, with the chatbot predicting what the next word should be in a sentence. 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.

And we’re doing this work leveraging capabilities from Nvidia and Databricks, two of the biggest names in AI today. We’re already seeing strong results, and we’re excited to showcase our work during the Nvidia GTC 2026 conference.

The most popular chatbots today, such as Claude and ChatGPT, are built using large language models, or LLMs, which are trained on huge amounts of unstructured data, such as text, video and photos. Our new foundation model is a different kind of deep learning neural network, called a large tabular model, or LTM, which is trained on structured data, such as large-scale tables or datasets.

We are training the latest version of our LTM on billions of anonymized transactions. Our plan is to ramp up this work to include hundreds of billions of payments transactions, as well as additional types of datasets, including merchant location data, fraud data, authorization data, chargeback data and loyalty program data. This work is powered by Nvidia’s advanced accelerated computing platform. By leveraging Nvidia's full-stack accelerated AI platform, we are able to process this data at unprecedented speeds.

As we train the model on more data and more kinds of data, it will be able to provide more insights and predict future transactions with greater accuracy.

One of the first areas we’re focusing on is cybersecurity. Our company has already built many of the best cyber tools in the industry to make commerce more secure. We believe adding this new foundation model’s capabilities into our current tools will make them even stronger.

To build our existing cybersecurity AI models, our data scientists start with raw transaction data. They then enrich this data with additional features to indicate what those models should analyze and flag. For example, a data scientist might add a feature that helps our AI models identify a sudden spike in someone’s purchase activity, which enables the model to detect and stop fraud.

In comparison, our new foundation model analyzes the same data with very limited human input as a starting point, learning more independently what the important characteristics of the data are. In this way, the LTM could identify new connections in the data that a human might not find on their own.

In our testing, we’ve already seen this new model outperform standard industry machine learning techniques, giving us promising early signs. For instance, very expensive but very infrequent purchases — such as when someone buys a wedding ring — tend to trigger current models today and cause a lot of false positives. In our experiments, our foundation model can better identify these legitimate transactions, with the model able to learn from relatively weak signals in the data.

We plan to build hybrid cybersecurity systems that combine the best of both our current AI models and this new LTM. This should help us build up and futureproof our cyber defenses.

This cybersecurity example is just one potential outcome of this research. We believe the new foundation model can also be used to improve loyalty and rewards programs, personalization models, portfolio optimization and data analytics tools.

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. This new LTM could become flexible enough to help us cut down significantly on having to maintain so many different models.

We’re ramping up our work to expand our LTM’s capabilities. We are looking to add algorithmic sophistication to the model’s architecture so it can gain more insights from raw data. Plus, we’re developing APIs and toolkits to give teams across Mastercard access to this new foundation model, so they can build new applications on top of it.

We will also continue our close collaboration with Nvidia and Databricks to push forward all this development work.

As always, we will continue building this new model following our data responsibility principles, focusing on user privacy, robust governance and controls and transparency.

With each step forward, just like the rapid development of chatbots today, we are starting to see the potential this foundation model could have for our industry — bringing more intelligence, security and speed into payments and commerce.