Governing Credit with Digital Twins: Explainable AI, Credit Officer Digital Twins, and the EU AI Act

Abstract

The deployment of artificial intelligence in credit decisioning has substantially improved predictive accuracy, yet the opacity of advanced machine learning models raises critical concerns about transparency, fairness, and regulatory compliance. The paper examines the intersection of Explainable AI (XAI), digital twin technology, and the European Union AI Act within the domain of credit origination and underwriting. A conceptual framework is proposed in which a digital twin of the credit deliberation process is complemented by a behavioural Credit Officer Digital Twin, together forming an integrated environment for real-time simulation, multi-level explainability, bias and efficiency monitoring, and regulatory audit. Drawing on recent literature, the paper synthesizes findings on SHAP, LIME, counterfactual explanations, fairness-constrained optimization, and financial and organizational digital twins to argue that twin-enabled XAI architectures can reconcile the tension between model performance, human judgement, and regulatory transparency. The Credit Officer Digital Twin is modelled as a data-driven replica of the human credit officer’s decision policy, augmented with analytics that surface individual biases, efficiency patterns, and inconsistencies, thereby supporting meaningful human oversight rather than replacing it. The overall framework operationalizes key requirements of the AI Act for high-risk AI systems, including data governance, documentation, human oversight, and the right to explanation under both the AI Act and the GDPR. Implications for banking practice, supervisory policy, and future research on responsible, human-in-the-loop credit decisioning are discussed. UDC: [004.8:336.77]+[341.1(EU):004.8]; JEL: E51, O33, C53