Mastercard Rebuilds Fraud Defences for the Age of Autonomous AI Shopping
Mastercard is reworking the fraud-detection systems underpinning its global payments network as artificial intelligence agents begin moving beyond product recommendations to making purchases directly on behalf of consumers.
The shift presents an unusual challenge for the payments industry. Automated software has historically been treated as a potential threat, with fraud engines trained to detect bots, block suspicious activity and prevent account takeovers. Agentic commerce now requires those systems to distinguish between malicious automation and authorised AI agents legitimately acting on a customer’s instructions.
Mastercard is addressing the problem through its Agent Pay infrastructure, which allows verified AI agents to initiate and complete payments using tokenised credentials rather than exposing a customer’s actual card details.
The framework is designed to provide merchants and banks with evidence that an agent is genuine, that the customer authorised the transaction and that the purchase remained within predetermined limits. This could include conditions covering the type of product, maximum price, approved merchant or timing of the transaction.
Central to the model are Mastercard Agentic Tokens, which create a traceable digital identity for the purchasing agent and connect each payment to the consumer’s verified intent. The company is also developing common protocols with technology platforms and payments partners to ensure that AI agents can interact safely with merchants across different systems.
The technology is intended to preserve accountability throughout the purchasing process, giving financial institutions clearer information for fraud investigations, chargebacks and dispute resolution when transactions are initiated by machines rather than directly by cardholders.
The redesign reflects a broader transformation in online commerce. Instead of manually searching websites, comparing prices and completing checkout forms, consumers could increasingly assign those tasks to AI assistants capable of finding products and executing purchases within agreed parameters.
For merchants, however, the rise of autonomous shoppers creates new risks. Retail systems must be able to identify legitimate agents while rejecting impersonators, malicious bots and automated tools attempting to exploit stolen payment credentials.
Mastercard’s approach combines tokenisation, transaction-risk analysis, agent verification and cryptographic proof of customer intent. The company is effectively adapting decades of fraud intelligence to a market in which automated behaviour can represent either an attack or a genuine customer transaction.
The payments group expects trust and interoperability to determine how quickly autonomous shopping gains widespread adoption. Consumers may be willing to delegate routine purchases to AI, but only if they retain control over spending and can clearly understand who authorised each transaction and how it was completed.