AI Fraud Detection, Privacy Safeguards, and Continued Use of Mobile Financial Services in Bangladesh

Artificial intelligence is increasingly being incorporated into financial-sector risk management, including automated anomaly detection and fraud monitoring. Yet the legitimacy of AI-enabled fraud controls depends not only on technical security gains but also on whether users perceive monitoring as proportionate, privacy-protective, reviewable, and institutionally accountable. This study develops a security-privacy-continuance perspective for mobile financial services (MFS) in Bangladesh and examines how AI literacy, perceived regulatory protection, and procedural fairness relate to users’ acceptance of AI-enabled fraud controls and continued-use intentions. The study follows the Group 8 research design and uses the Q8 five-point Likert questionnaire. The supplied database contains 385 records; however, its Codebook identifies the respondent sequence and contextual comments as synthetic, so the empirical results are explicitly treated as demonstrative rather than field evidence. Composite reliability diagnostics reveal a major measurement problem: Cronbach’s alpha is low for AI literacy (α=.322), fairness (α=.193), regulatory protection (α=.250), and the full fraud/privacy/continuance block (α=.224). This finding is substantively important rather than merely technical: the instrument’s items appear to combine related but non-identical governance dimensions, making a single latent construct inappropriate without refinement. Item-level robust regressions nevertheless show that regulatory protection is positively associated with both continued-use indicators (FPC4: b=.335, p=.002; FPC5: b=.255, p=.009), while AI literacy is positively associated with perceived fraud-detection benefit (FPC1: b=.242, p=.019) and false-alert reviewability (FPC3: b=.267, p=.015). In the aggregate model, AI literacy, fairness, and regulatory protection jointly explain 15.8% of variance in the FPC composite. The study’s principal contribution is therefore a measurement-aware account of AI financial trust: users may support fraud detection while simultaneously requiring privacy safeguards, review mechanisms, and credible regulatory protection. For Bangladesh, the implication is that AI fraud governance should be designed as a socio-technical consumer-protection architecture rather than as surveillance technology alone. The manuscript identifies precise questionnaire revisions and a validation pathway required before substantive publication.