Shifting Audits Left: A Framework for Enhancing AI Auditability from an Internal Auditing Lens

Abstract

Artificial intelligence (AI) systems have become integrated into our lives more than ever. However, AI systems do not come without drawbacks. Over recent years, these systems have demonstrated embedded discrimination and bias, which have shaken public trust in these systems. To address these risks, auditing AI systems has emerged as a core process to regain stakeholder trust by identifying AI risks and planning mitigation strategies to address them. Despite its promising impact, this field is still immature, lacks standards and well-defined practices, and practitioners often face substantial challenges in anticipating, detecting, and mitigating harms. To address these challenges, the article highlights the importance of AI systems’ auditability and the major role internal auditing plays in enhancing this auditability. We then present an innovative framework aimed at improving the auditability of AI systems through the process of internal auditing. The proposed framework is intended to serve as an auditability enhancer and enabler to address AI harms before they propagate to the end user.