Audit Program on Artificial Intelligence (AI)-driven Credit Risk Models

Abstract

From an Internal Audit perspective, the integration of Artificial Intelligence (AI) into credit risk modelling through Machine Learning (ML) algorithms presents significant challenges due to the complexity and multidimensional nature of these models. While AI enhances predictive performance and accuracy, its inherent lack of transparency and explainability increases the risk of control deficiencies, potentially leading to financial losses, misrepresentation of information, unfair discrimination against debtors, and non compliance with EU regulations. This paper introduces a comprehensive audit framework designed to establish robust internal controls over AI-driven credit risk models. Aligned with the Model Risk Management (MRM) lifecycle, we propose a structured set of audit tests and controls, organized by thematic area, to assess key aspects such as model design and performance, governance, reliability, and regulatory compliance. Additionally, we provide practical examples in emerging areas to illustrate their application. These audit procedures aim to identify critical vulnerabilities while ensuring adherence to regulatory standards, including EBA/REP/2023/28 and the evolving requirements of the EU AI Act.

Keywords

• internal audit
• credit risk
• artificial intelligence
• eu regulation
• model risk