Toward Secure and Trustworthy Artificial Intelligence in Financial Data Systems: A Governance and Risk-Control Framework

Abstract

This study examined the problem of fragmented, inconsistent, and after-the-fact controls in the use of artificial intelligence within financial data systems, where AI models supporting credit, fraud, trading, and compliance decisions are exposed to adversarial manipulation, data leakage, bias, and opacity, and where governance, when present, is often disconnected from technical enforcement. The purpose of the study was to assess how a governance and risk-control framework, integrating data security and privacy controls, model robustness and resilience, transparency and explainability, governance and accountability, and human oversight, influences trustworthy AI assurance in financial data systems. A quantitative, cross-sectional, case-based design was adopted, and data were collected through a structured five-point Likert-scale questionnaire from 146 valid respondents out of 165 distributed questionnaires, representing an 88.5% valid response rate. The sample included risk and compliance officers, data scientists and machine-learning engineers, AI and model-governance leads, security and privacy engineers, and IT auditors and regulators, with 67.8% directly involved in AI governance, security, or risk-control activities. The key variables were data security and privacy controls, model robustness and resilience, transparency and explainability, governance and accountability, human oversight and institutional trust, framework design quality, and trustworthy AI assurance. The analysis plan included descriptive statistics, reliability testing using Cronbach's alpha, Pearson correlation, regression modeling, a framework maturity index, and an AI risk-control priority matrix. The headline findings showed that all major constructs were rated high, with trustworthy AI assurance recording the highest mean score of 4.23, followed by governance and accountability at 4.16 and framework design quality at 4.12. Reliability was acceptable to excellent, with Cronbach's alpha values ranging from 0.82 to 0.93. Correlation results showed significant positive relationships, including r = 0.79 between framework design quality and trustworthy AI assurance. Regression results confirmed that the model explained 69.3% of the variance in trustworthy AI assurance, R² = 0.693, adjusted R² = 0.680, F(6,139) = 52.28, p < 0.001. Framework design quality was the strongest predictor, β = 0.33, followed by governance and accountability, β = 0.25, and data security and privacy controls, β = 0.22. The findings imply that financial institutions should strengthen adversarial robustness, privacy protection, model interpretability, continuous monitoring, and human oversight to improve the security, transparency, and trustworthiness of AI in financial data systems.