AI Fights Fraud: Mastercard's Millisecond Defense

Mastercard is enhancing its fraud protection capabilities with sophisticated AI models, particularly through its Decision Intelligence Pro (DI Pro) platform. This system operates at immense scale, analyzing around 160 billion transactions annually, with peak rates reaching 70,000 per second. DI Pro works by assessing the risk of each transaction in real-time, a process that typically concludes within 300 milliseconds before the issuing bank makes a final decision.

The core of DI Pro is a recurrent neural network with an 'inverse recommender' architecture. This approach models fraud detection as a recommendation challenge, enabling the AI to understand how merchants relate to each other and identify deviations from typical user behavior. Mastercard also employs unique data sovereignty strategies, using aggregated, anonymized data to train global models without compromising local privacy regulations.

In an ongoing battle against evolving fraud tactics, Mastercard is actively engaging cyber criminals. The company deploys 'honeypots,' artificial environments designed to trap malicious actors. By interacting with detected threats, AI agents aim to uncover mule accounts used for illicit fund transfers. This strategy allows for the mapping of global fraud networks by identifying connections between mule accounts and legitimate financial channels.