Intelligent Audit Systems: A Systematic Review of Fundamental Scientific Challenges

Abstract

The integration of Artificial Intelligence, Blockchain, and Process Mining into auditing systems promises improvements in automation, transparency, and analytical capability. However, fundamental scientific challenges remain unresolved. This paper presents a systematic literature review of 37 peer-reviewed studies (2023–2026) to identify three structural barriers limiting the development of truly intelligent audit systems: (1) the validity and probative value of digital traces, (2) the faithful reconstruction of real-world processes from imperfect logs, and (3) the detection of complex, rare, and adaptive deviations. Our analysis shows that despite advances in federated learning, deep learning–based anomaly detection, and blockchain audit trails, none fully addresses the foundational data validity problem. The classical "garbage in, garbage out" principle persists across all technological perspectives. Human expertise remains essential for contextual validation, and a persistent trade-off between predictive performance, explainability, and adaptability constrains all current approaches.