Zero Trust Cybersecurity Using Generative AI & Multi-Agent Systems for Cloud-Native Enterprises

Abstract

The fast adoption of cloud-native solutions has contributed to the complexity of cybersecurity threats in distributed enterprise environments. Perimeter-based cybersecurity approaches have proven ineffective in protecting cloud workloads, identities, APIs, and networks from attacks. In this research paper, the effectiveness of integrating Zero Trust Cybersecurity principles with Generative Artificial Intelligence and Multi-Agent Systems for better security in cloud-native organisations is analysed. The quantitative experimental research methodology was used with the simulated data set of 5,000 cloud security events, which comprised legitimate and malicious access attempts. The process of data preprocessing involved cleaning, normalising, feature encoding, and class balancing if necessary. Three machine learning models – Random Forest, XGBoost, and a Generative AI-based model for threat detection – were created for classification of cloud security events. Multi-Agent systems were designed for the monitoring of identity verification, network traffic, workloads, API activities, and policy enforcement in real-time. The findings revealed that XGBoost demonstrated the highest classification accuracy (99.1%) with an AUC-ROC score of 0.999, and the Generative AI threat detection model scored a 97.06% F1 score and 99.10% threat detection accuracy. The Multi-Agent System delivered 98.5% security event reduction efficiency with an average time delay of 177 milliseconds, which made real-time monitoring of the cloud environment. Feature importance evaluation showed that IP reputation value, user trust score, and device validation were the most important factors for predicting malicious access attempts. These results show that the integration of the Zero Trust concept, Generative AI technologies, and security agents improves continuous authentication, intelligent threat detection, and automatic response to security incidents. However, the research is confined by the use of artificial data.