TAIMS: Transaction Artificial Intelligence Monitoring Systems Framework

Sammanfattning

Money laundering encompasses a crucial risk to Financial Institutions (FIs), requiring them to adhere to Anti-money Laundering (AML) practices such as Transaction Monitoring (TM). Recently, increasing attention has been given in applying Artificial Intelligence (AI) and Machine Learning (ML) to improve the efficiency of Transaction Monitoring Systems (TMS). However, the integration of these technologies introduce ethical and regulatory challenges. To manage such risks, Gartner proposed the Artificial Intelligence Trust, Risk and Security Management (AI TRiSM) framework, offering a structured approach to responsible AI/ML use. Additionally, the EU Artificial Intelligence Act (AIA), effective August 1, 2024, establishes the first legal framework for ethical AI/ML deployment. Despite these advancements, limited research exists on how FIs can ethically implement AI/ML in TMS, particularly in light of the evolving regulatory landscape. This thesis aimed to fill this gap by developing a strategic framework for implementing AI/ML ethically in TMS by combining the AI TRiSM framework, the AIA and insights from a FI by answering the following research question: How can FIs implement AI/ML in TMS while upholding ethical principles? Through using a mixed qualitative method strategy rooted in Design Science Research (DSR) and Case Study Research, the study followed a five-step iterative process: 1) Explicating the Problem (Stakeholder Deliberations; Systematic Literature Review (SLR)), 2) Defining Requirements (Semi-structured Interviews with case study company stakeholders; Document Analysis; Thematic Analysis), 3) Designing and Developing the Artifact (Brainstorming; Sketching), 4) Demonstrating the Artifact (Fictional Case in Real Life Setting) and 5) Evaluating the Artifact (Naturalistic Formative ex ante Evaluation through a Focus Group session with case study company stakeholders). This resulted in the Transaction Artificial Intelligence Monitoring Systems (TAIMS) Framework. The findings from the evaluation of the TAIMS framework revealed strong stakeholder agreement on its suitability and comprehensive nature for AI/ML implementation in TMS, with particular praise for its ethical risk management, governance and flexibility. However, stakeholders highlight areas for improvement in addressing the framework’s technical complexity, incorporating responsibilities of the providers and clearer guidance on human oversight of the AI/ML system to ensure ethical AI/ML use. Future work could include testing the TAIMS framework in real-world TMS, applying it to other high-risk AI/ML areas, refining it further with the provider perspective and continuously monitoring the AIA.