Agentic AI Security in 6G Networks: A Survey of Emerging Attack Vectors, Vulnerabilities, and Defenses

Authors:
  • HFHailin Feng
    Hailin Feng
  • YXYuchao Xia
    Yuchao Xia
  • YZYongkang Zhao
    Yongkang Zhao
  • ZWZhihao Wen
    Zhihao Wen
  • JCJijing Cai
    Jijing Cai
  • PBPronaya Bhattacharya
    Pronaya Bhattacharya
  • KFKai Fang
    Kai Fang
  • MLMadhusanka Liyanage
    Madhusanka Liyanage

Abstract
The emergence of autonomous, goal-driven AI agents as core components of future 6G networks fundamentally reshapes the network security landscape. Unlike earlier AI-assisted systems, these agents operate with a high degree of autonomy, enabling adaptive decision-making that current security models are not designed to handle. In this survey, we examine how agentic AI in 6G can be exploited by attackers and, conversely, how these AI systems themselves become vulnerable. We discuss how malicious actors can leverage autonomous agents to conduct intelligent, evolving attacks across the radio access network, network control functions, and digital twin environments, while also showing how the agents' own capabilities for reasoning, memory, and tool use introduce new vulnerabilities to poisoning, hijacking, and manipulation. By integrating these two perspectives, we assess emerging defense strategies, identify gaps in existing standards and governance, and outline key research directions toward building trustworthy, AI-native 6G networks. Our goal is to lay a practical groundwork for securing future intelligent networks.