Agentic Personal Finance Advisors with Privacy-Preserving Decision Logs: Federated Agents That Explain and Never Leak

Abstract

The evolution of agentic artificial intelligence (AI) in financial technology is redefining how personal finance advisory systems operate, bridging autonomy, transparency, and privacy. This paper introduces Agentic Personal Finance Advisors with Privacy-Preserving Decision Logs, a federated agent framework that explains financial decisions without compromising user confidentiality. Unlike conventional centralized advisory systems that risk data exposure, the proposed architecture leverages federated learning to train distributed models across user devices while maintaining data sovereignty. Each agent operates autonomously to deliver personalized investment, budgeting, and risk assessment recommendations, guided by explainable decision protocols inspired by multi-agent interpretability mechanisms. Privacy-preserving decision logs employ differential gradient masking and encrypted federated aggregation to mitigate inference and gradient-leak attacks identified in prior studies. By integrating agentic behavior with human-centered explainability, the system ensures compliance with regulatory intelligence standards while enhancing user trust. This research contributes a novel federated-agentic architecture for finance that “explains and never leaks,” ensuring adaptive, secure, and interpretable financial guidance. Empirical evaluations demonstrate improved F1-metrics for decision reliability and significant reductions in privacy leakage across federated environments.