Authors:
S. Vishal, S. Sai Vishaal

Addresses:
Department of Artificial Intelligence and Machine Learning, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. 

Abstract:

Zero-Trust Architectures (ZTA) migration involves transitioning from perimeter-based security to persistent authentication for every request to access resources. Although it introduces security, the migration comes with a high overhead cost, inducing network latency and increased complexity in policy administration. This paper presents an AI-based framework for ZTA that minimizes network traffic flow without compromising security tenets. Our solution involves a deep reinforcement learning (DRL) agent that dynamically varies network paths and access privileges in real-time depending on device posture, end-user behaviour, and app sensitivity. The study was proven in an emulated enterprise network deployment. The primary data used here is the 'ZTA-Traffic-Sim-2025', which contains simulated data of 10 million network flow records. Traffic for 5,000 users and 15,000 devices was simulated for a month, incorporating various simulated attack vectors. The model is trained using Python with TensorFlow for the DRL agent and the ns-3 simulator to simulate the network environment. The results confirm that the AI-based ZTA reduces average network latency by up to 35% and throughput by 25% compared to static ZTA, while improving the detection rate for anomaly activity by 18%.

Keywords: Zero-Trust Architecture; Artificial Intelligence; Network Traffic Optimization; Deep Reinforcement Learning; Device Posture; End-User Behaviour; App Sensitivity; Network Environment.

Received on: 03/09/2024, Revised on: 15/11/2024, Accepted on: 17/12/2024, Published on: 03/06/2025

DOI: 10.69888/FTSIN.2025.000380

FMDB Transactions on Sustainable Intelligent Networks, 2025 Vol. 2 No. 2, Pages: 59-68

• Views : 57
• Downloads : 9

Download PDF