Securing Next-Generation Wireless Networks Against Native GenAI Attacks: An Evidence-Theoretic Approach
Md. Shirajum Munir, Sravanthi Proddatoori, Manjushree Muralidhara, Marco Gamarra, Walid Saad, Zhu Han, Sachin S. Shetty · 2025
Intelligent poisoning attacks will pose fundamental challenges for sixth-generation (6G) wireless network security due to the massive deployment of native AI in radio units as well as in core networks. Network metrics and parameters, which are inherently uncertain, can become susceptible to intelligent poisoning through native generative AI (GenAI) mechanisms. In this paper, GenAI-driven intelligent attacks in wireless networks are investigated in order to understand their impact and severity by using uncertainty-informed root cause analysis. Then, a new approach for mitigating GenAI-driven attacks is proposed through the use of trustworthy service aggregation. First, a joint decision problem is formulated for generating intelligent adversarial attacks, understanding uncertain attack severity, and mitigating them in wireless networks. Second, a novel evidencetheoretic trustworthy AI (ET-TAI) framework is developed to address the formulated problem by understanding the root-cause of the native GenAI-driven intelligent attack and establishing defense in wireless networks. In particular, the proposed ET-TAI framework enables a narrow GenAI scheme that is designed to penetrate intelligent adversarial attacks in wireless networks’ metrics and parameters. Then a Dempster–Shafer-based mechanism that is deployed to capture the uncertain behavior of those intelligent attacks through prior evidence to quantify the trust for further mitigation. Extensive experimental analysis shows the proposed ET-TAI framework’s efficacy in understanding the trust in GenAI-driven intelligent poisoning attacks on network parameters and metrics by quantifying root causes and mitigating rates. Results show that the GenAI can penetrate intelligent poisoning attacks with high reconstruction capabilities of 95% for downlink services.