On the Auditor’s Verification of Accounting Estimates Generated by Artificial Intelligence Systems

This article addresses the verification of accounting estimates generated by artificial intelligence systems. It investigates the «black box» problem in machine learning and substantiates the need to integrate the concept of Explainable Artificial Intelligence (XAI) into modern auditing practices to enhance audit quality. The research methodology includes the analysis of regulatory frameworks, a comparative analysis of traditional and algorithmic audit procedures, and scientific modeling. The study demonstrates that traditional audit methods are ineffective for neural network-based estimates due to their fundamental opacity and lack of explicit causal relationships. Applying XAI enables auditors to verify the economic soundness of model weight coefficients and input parameters. The article proposes a structured algorithm of audit procedures for verifying AI-generated accounting estimates. Comprising four stages, this algorithm integrates the requirements of International Standards on Auditing with modern XAI techniques. These procedures encompass the preliminary understanding of the AI system, verification of input data, application of XAI methods to validate the estimates, and the formulation of the audit report with recommendations.