The Role and Challenges of Artificial Intelligence in Mitigating Fraud Risk in Digital Wallet Use

Purpose: This study examines the effectiveness of Artificial Intelligence (AI) in detecting fraud in digital financial transactions compared with manual methods from an internal control perspective, identifies its limitations, particularly false positives and machine learning blind spots, and explains the role of system calibration and human intervention. 
Research Methodology: The study used a literature review of scientific articles, professional reports, and regulations on AI based fraud detection in digital financial transactions, including digital wallets. The literature was screened for relevance to the research questions and synthesized thematically. 
Results: AI outperforms manual and rule based methods in speed, scale, and detection accuracy because it can analyze entire transaction populations in real time. However, AI has structural limitations in the form of false positives and blind spots toward new fraud patterns caused by concept drift, data bias, adversarial behavior, and limited labeled data. Regular calibration and human professional judgment remain necessary to verify alerts and make final decisions, in line with regulations that require human involvement in automated decisions. 
Conclusions: Collaboration between AI and humans, with AI as the initial screening tool and humans as the final decision makers, is the most effective model for fraud detection. 
Limitations: The study is a literature review and has not been tested empirically with real transaction data. 
Contributions: The study integrates evidence on AI effectiveness and limitations into a human in the loop framework and offers guidance for financial institutions, regulators, and researchers.