Downloads: 1
Research Paper | Computer Science | Volume 15 Issue 3, March 2026 | Pages: 337 - 343 | India
AI-Based Anomaly Detection in Air-Gapped Environments
Abstract: Air-gapped systems are widely used to secure critical infrastructures such as military networks, nuclear facilities, and government information systems by maintaining strict physical isolation from external networks. Despite this isolation, such systems remain vulnerable to sophisticated cyber threats introduced through removable media, insider activities, or hardware manipulation. Traditional security mechanisms relying on signature-based detection are often ineffective in identifying unknown or stealthy attacks in these environments. This paper proposes an Artificial Intelligence-based anomaly detection framework specifically designed for air-gapped systems. The approach employs a hybrid unsupervised learning architecture combining Autoencoder neural networks and Isolation Forest algorithms to model normal system behaviour and detect deviations that may indicate malicious activity. The system continuously analyzes host-level telemetry including CPU usage, memory access patterns, process execution, file operations, and USB device interactions. Experimental evaluation in a simulated air-gapped environment demonstrates that the proposed system achieves detection accuracy of approximately 96% with a low false-positive rate. The results indicate that behavior-based anomaly detection using hybrid machine learning techniques can provide an effective and autonomous security mechanism for protecting critical air-gapped infrastructures against emerging cyber threats.
Keywords: Air-Gapped Cybersecurity, Behavior Anomaly Detection, Autoencoder Neural Networks, Isolation Forest, Critical Infrastructure Security, Host Based Intrusion Detection
How to Cite?: Dr Abhishek Kumar, Meenakshi Saini, "AI-Based Anomaly Detection in Air-Gapped Environments", Volume 15 Issue 3, March 2026, International Journal of Science and Research (IJSR), Pages: 337-343, https://www.ijsr.net/getabstract.php?paperid=SR26308141537, DOI: https://dx.dx.doi.org/10.21275/SR26308141537