Mastercard rolls out AI “Agent Detection Score” aimed at flagging risk in agent-initiated payments

Mastercard said it has unveiled an AI-powered “Agent Detection Score” designed to help payment processors and merchants assess the risk of transactions that may be initiated by software agents, such as automated systems acting on behalf of users. The announcement comes as payments infrastructure tries to adapt to a growing range of digital workflows, including increasingly automated purchasing and account activity driven by external tools.

According to Mastercard, the score is part of what it calls its Agent Pay Trust Framework. Within that framework, the Agent Detection Score assigns a probability intended to capture whether a given payment was initiated by an AI agent. The company’s aim is to provide a consistent risk announcement that can be used downstream in decision-making, such as routing, approvals, or additional verification steps.

The company positioned the new score as a targeted enhancement rather than a general-purpose fraud tool. The focus is specifically on “agent-initiated” transaction risk, which Mastercard treats as a distinct category from more traditional payment fraud patterns. That distinction matters because automated agents can behave differently from human users, potentially making standard detection methods less effective if they are not trained or tuned for agent-like transaction behavior.

Mastercard did not provide in the announcement how the probability is calculated, what data features are used, or how the score is calibrated across different merchant types and geographic markets. It also did not disclose performance metrics such as detection accuracy, false-positive rates, or how quickly the scoring model can be updated as new agent behaviors emerge.

Still, Mastercard’s move reflects a broader industry shift toward machine-learning risk scoring that can be applied at high speed inside payments networks. Agent-initiated activity is expected to grow as more businesses integrate automation for purchasing, billing, and customer service workflows. As that happens, the payment ecosystem faces a challenge: separating legitimate automation from misuse, including scenarios where an AI system is used to attempt fraudulent transactions or to scale up attacks.

Mastercard framed its work as part of a trust architecture for Agent Pay, suggesting the company views risk management as an ecosystem function, not a single model deployed in isolation. By bundling the Agent Detection Score into a named framework, the company is indicating that it expects partners to consume the score as an input to broader trust and compliance processes.

The company also did not specify whether the Agent Detection Score is available immediately to all partners, whether it is delivered as a new API capability, or how partners will integrate it into existing decision systems. Nor did Mastercard detail whether the score is intended for specific transaction types, such as particular payment channels or authorization flows. Those implementation details will be critical for assessing how quickly the score can be operationalized.