AN AGENTIC AI ARCHITECTURE FOR NATIONAL-SCALE FRAUD DETECTION IN DIGITAL PAYMENTS ECOSYSTEMS

Real-time national fraud detection platforms — exemplified by Brazil's PIX, India's UPI and Nigeria's NIBSS — embed centralised AI-and-rule-driven monitoring at the settlement layer, yet share three structural limitations: human-in-the-loop investigation bottlenecks, batch retraining latency and weak cross-institution coordination. This paper proposes an architectural successor: an agentic AI fabric in which autonomous agents, with bounded autonomy, persistent memory and observable handoffs, distribute detection, investigation, remediation and learning across the terminal, institutional and national-authority layers. A worked example anchored on the NIBSS National Payment Stack in Nigeria, under the Central Bank of Nigeria's January 2025 fraud-debit directive, illustrates the design end-to-end. Extensibility to AML, KYC/KYB, consumer protection and cross-border supervision is discussed, with the principal risks of model governance, accountability and regulatory acceptance. The proposition aligns with the BIS Innovation Hub's emerging SupTech direction.