AI-Based Deepfake Detection for Fraud Prevention and Cybersecurity in U.S. Digital Banking Systems

Purpose: The authors examine the impact and contributions of AI solutions to detect deepfakes on fraud prevention and cybersecurity measures in digital banking in the USA. Specifically, it explored the relationship between AI-Driven Multimodal Deepfake Detection Capability (AMMDFC) on Digital Banking Fraud and Cybersecurity Resilience (DBFCR) and the relationship between AI-Enabled Real-Time Deepfake Detection Capability (ARTDDC) on DBFCR and between Explainable AI-Based Deepfake Detection Capability (XAIDDC) and DBFCR, with Digital Banking Fraud and Cybersecurity Resilience (DBFCR) serving as a mediator. Design/Methodology/Approach: Quantitative and cross-sectional design were used. 510 bankers, cyber experts, fraud prevention, risk, IT and digital-finance professionals were sampled. The proposed models in terms of measurement and structure were analysed in a PLS-SEM application with a bootstrapping process via the SmartPLS software program version 4.0. The proposed models regarding the measurement and structure were tested with the PLS-SEM with 1,000 bootstrapping, using the software program SmartPLS. Findings: First, the results indicated that AMMDFC, ARTDDC, and XAIDDC were found to have a positive and significant effect on DTDE and DBFCR. DTDE also managed to make significant improvements for DBFCR. The model was successful in explaining 54.2% of the mean variation for DE of DT and 64.0% of the mean variation for DBFCR. Second, the three descriptions of AI were fully mediated through DTDE in terms of relation to DBFCR. Secondly, three descriptions of AI were linked to DBFCR all the way through DTDE. DTDE was partially responsible for all three AI skills and DBFCR. ARTDDC was more effective on DTDE, highlighting the significance of timely detection in the digital banking environment. Theoretical Implications: The study statistically anchored the Resource-Based View, Dynamic Capabilities Theory, Organisational Information Processing Theory and Organisational Resilience Theory and also highlighted the translation of AI capabilities into cybersecurity, mediated by effective threat detection. Practical Implications: Banks need to incorporate multimodal analysis, real-time detection capabilities, explainable AI features, human controls, and incident response protocols into their technology to improve their defences against synthetic identities, voice cloning, and fabricated digital content.