Research on Algorithmic Explainability and Legal Liability Framework under AI Audit Background

Abstract

The deep integration of artificial intelligence (AI) in auditing has significantly enhanced audit efficiency and data processing capabilities. However, the inherent “black box” nature of algorithms poses severe challenges in determining liability for audit failures, which hinders further development of AI auditing technologies. This study aims to address liability attribution dilemmas caused by algorithmic opacity in AI auditing contexts. By integrating algorithm governance theory with extended audit accountability theory, we establish a bidirectional framework of “algorithmic explainability and legal liability.” The research first analyzes the real-world contradiction between explainability deficiencies and legal lag in current AI auditing practices, revealing Deloitte's liability determination challenges stemming from generative AI hallucinations. Solutions are then proposed from both technical and legal dimensions. Findings indicate that algorithmic explainability serves as a prerequisite for legal liability determination, while clear legal regulations provide institutional safeguards for technological transparency. This framework not only clarifies accountability chains in human-machine collaboration but also offers theoretical support and practical guidance for promoting compliance in AI auditing and maintaining capital market trust mechanisms.