Abstract

The paper presents a systematic literature review of 16 studies (2010–2025) comparing AI models with rule-based systems and human analysts for insider-threat detection in terms of accuracy and speed. AI approaches generally achieve over 90% detection accuracy and sometimes near real-time performance, especially with sequential deep learning and probabilistic models, but face issues such as imbalanced datasets, computational cost, and limited interpretability. Overall, AI-based methods outperform traditional techniques, yet no single model resolves all trade-offs, highlighting the need for hybrid solutions that combine AI efficiency with human judgment and rule-based safeguards.