Agentic Compliance-by-Design: An Interpretable Agent Architecture for Real-Time AML/KYC Actions in Fintech Platforms

Abstract

Agentic Compliance-by-Design presents an interpretable, modular agent architecture for enabling real-time Anti-Money Laundering (AML) and Know-Your-Customer (KYC) actions within fintech platforms. The paper introduces agents that integrate rule-based legal constraints, probabilistic risk scoring, and ledger-aware verification to execute, explain, and audit intervention decisions with low latency. The architecture combines a compliance orchestration layer that encodes statutory and policy constraints, a trust layer for cryptographic provenance and smart-law triggers, and an explainability module that exposes causal rationales to auditors and regulators. We validate the approach using simulated transaction streams and SME banking scenarios, demonstrating high detection recall while preserving transaction throughput and offering concise human-readable justifications for automated holds, escalations, and customer re-onboarding. The design leverages principles from smart-law and blockchain compliance to reconcile programmable rules with regulatory intent, enabling adaptive governance and policy versioning. By embedding interpretability and auditability at the agent level, the system reduces operational burden, supports regulatory examinations, and facilitates lawful automation across jurisdictional regimes. We situate our contributions against contemporary frameworks for law-following AI and unified compliance intelligence. We extend key work on blockchain compliance, law-following AI,  smart-law, and unified compliance.