DFAS-AML: An Integrated Conceptual Governance Framework for Ethical, Explainable, and Sovereign-Sensitive AI in Anti-Money Laundering Systems

Hasan Alaali (DFAS Research Center)

Journal of Governance Risk Management Compliance and Sustainability
Vol. 6 No. 2 (2026): October Volume

Artificial intelligence is increasingly embedded in anti-money laundering (AML) systems, yet improved analytical capability does not by itself ensure explainability, accountable human intervention, ethical resilience, jurisdiction-sensitive processing, or decision traceability. Existing research and regulatory instruments address these concerns but remain fragmented across technical, legal, organizational, and data-governance domains. This study adopts a normative conceptual and design-oriented methodology. It synthesizes peer-reviewed research and authoritative regulatory instruments, translates identified governance problems into operational requirements, derives corresponding governance mechanisms, and evaluates the resulting architecture through conceptual coherence, functional completeness, traceability, regulatory relevance, sovereign adaptability, operational interpretability, and illustrative stress scenarios. The analysis identifies five interdependent governance requirements: explanation governance, accountable human override, ethical resilience, sovereign calibration, and controlled decision traceability. These requirements are integrated into the Dynamic Financial Applied Meta-Science for Anti-Money Laundering (DFAS-AML) through five coordinated components: A-AMLEEF, A-AMLOCC, A-AMLD, A-SCEAML, and A-AMIL. The principal analytical finding is that effective governance of AI-enabled AML requires governance of the complete decision lifecycle rather than the detection model alone, linking explanation adequacy, ethical and jurisdictional assessment, authorized intervention, institutional action, and tamper-evident decision reconstruction. DFAS-AML provides an AML-specific governance architecture that integrates functions otherwise addressed across separate governance domains. It complements rather than replaces existing AML and AI-governance instruments and does not establish empirical effectiveness. The architecture remains conceptually evaluated and requires subsequent expert, institutional, empirical, and regulatory validation.