About This Publication

A novel deep synthesis-based insider intrusion detection (DS-IID) model for malicious insiders and AI-generated threats is a 2025 Scientific Reports study addressing insider-threat detection in environments where generative AI can create synthetic user profiles and imitate legitimate behavior.

The authors introduce the Deep Synthesis Insider Intrusion Detection (DS-IID) model, combining deep feature synthesis, generative models, and binary deep learning to detect malicious insiders, evaluate how effectively generative algorithms reproduce real user profiles, and distinguish real abnormal behavior from AI-generated synthetic threats. The model is evaluated on the CERT r4.2 insider-threat dataset, which contains user logon, email, device, file, HTTP, and psychometric activity across 1,000 employees and more than 32 million events.

The study reports approximately 97% accuracy and an AUC of 0.99 for malicious-insider detection, while its synthetic-profile experiments show greater than 99% accuracy under selected generation conditions. The analysis also compares Gaussian Copula, CTGAN, TVAE, and CopulaGAN approaches for synthetic tabular data and examines the problem of realistic fabricated identities and behavior patterns evading traditional intrusion-detection systems.

The paper is particularly relevant to cybersecurity research involving insider threats, behavioral anomaly detection, synthetic identities, AI-generated user activity, deep learning, deep feature synthesis, generative adversarial techniques, and the development of adaptive detection systems for emerging AI-enabled threats.

Original Publication: January 2, 2025

Journal: Scientific Reports

Volume / Article: 15, Article 207 (2025)

DOI: 10.1038/s41598-024-84673-w

Publication Status: Peer-Reviewed, Open Access

License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Authors

• Hazem M. Kotb — The Institute of Cancer Research, London
• Tarek Gaber — University of Salford; Suez Canal University
• Salem AlJanah — IMSIU; University of Manchester
• Hossam M. Zawbaa — Beni-Suef University; Applied Science Private University
• Mohammed Alkhathami — Imam Mohammad Ibn Saud Islamic University

The article credits all five authors with conceptualization, methodology, software, validation, analysis, resources, original-draft preparation, review/editing, and visualization. Tarek Gaber and Salem AlJanah are identified as corresponding authors.