A GenAI-Driven Zero-Trust Cybersecurity Mesh for Real-Time Fraud Detection in Digital Payment Networks

Abstract

The rapid expansion of digital payment ecosystems has significantly increased the complexity and scale of financial fraud. Traditional centralized fraud detection engines struggle to provide real-time, context-aware risk assessment across distributed and API-driven infrastructures. Recent advances in Zero-Trust Architecture (ZTA) and cybersecurity mesh frameworks provide structural resilience yet lack adaptive contextual reasoning. This paper proposes a GenAI-Driven Zero-Trust Cybersecurity Mesh (GZTCM) designed for real-time fraud detection in high-throughput payment networks. The proposed architecture integrates distributed risk enforcement nodes with a generative AI–augmented contextual anomaly reasoning engine. A formal threat model is developed to quantify trust validation and probabilistic fraud scoring. The system is evaluated using a synthetic payment dataset reflecting realistic transaction distributions and adversarial patterns. Experimental results demonstrate improvements of 8.4% in F1-score and 21% reduction in false positives compared to conventional gradient boosting baselines, while maintaining sub-120ms inference latency. The findings indicate that embedding generative contextual reasoning within a zero-trust distributed mesh enhances both detection robustness and operational scalability. The proposed framework contributes computationally grounded architecture and empirical validation suitable for next-generation digital payment infrastructures.