Rethinking Secure Semantic Communications in the Age of Generative and Agentic AI: Threats and Opportunities

TL;DR

The paper reframes secure semantic communications in the GenAI and agentic-AI era, offering a taxonomy of eavesdropping threats based on attacker access and knowledge. It analyzes how GenAI models (GANs, diffusion models, LLMs) and agentic components (memory, external knowledge, reasoning, RAG, intent inference) can dramatically enhance private-information reconstruction and behavioral inferences. It then outlines privacy-preserving opportunities, including GenAI-assisted physical-layer security and covert communication, as well as agentic-context-aware and evolving security paradigms. Through case studies, the work demonstrates both the heightened risks from advanced eavesdropping and practical defenses that leverage GenAI and agentic AI to protect semantic privacy in SemCom systems. The findings highlight the need for new security metrics and adaptive, context-aware designs to ensure privacy in future 6G networks.

Abstract

Semantic communication (SemCom) improves communication efficiency by transmitting task-relevant information instead of raw bits and is expected to be a key technology for 6G networks. Recent advances in generative AI (GenAI) further enhance SemCom by enabling robust semantic encoding and decoding under limited channel conditions. However, these efficiency gains also introduce new security and privacy vulnerabilities. Due to the broadcast nature of wireless channels, eavesdroppers can also use powerful GenAI-based semantic decoders to recover private information from intercepted signals. Moreover, rapid advances in agentic AI enable eavesdroppers to perform long-term and adaptive inference through the integration of memory, external knowledge, and reasoning capabilities. This allows eavesdroppers to further infer user private behavior and intent beyond the transmitted content. Motivated by these emerging challenges, this paper comprehensively rethinks the security and privacy of SemCom systems in the age of generative and agentic AI. We first present a systematic taxonomy of eavesdropping threat models in SemCom systems. Then, we provide insights into how GenAI and agentic AI can enhance eavesdropping threats. Meanwhile, we also highlight potential opportunities for leveraging GenAI and agentic AI to design privacy-preserving SemCom systems.

Figures (5)

• Figure 1: Illustration of eavesdropping models in SemCom systems, which summarizes four representative threat models based on the eavesdropper’s access level and model knowledge, and categorizes representative existing technical papers into each threat category.
• Figure 2: Illustration of GenAI-enabled eavesdropping in SemCom systems, which shows how different types of GenAI models can be exploited by eavesdroppers to enhance private information reconstruction from intercepted semantic signals. We also present an agentic AI empowered eavesdropping framework, where the eavesdropper operates as an intelligent agent that continuously observes, reasons, memorizes, and retrieves knowledge from external knowledge bases to improve private information reconstruction and further infer user behavior and intent.
• Figure 3: Illustration of typical GenAI-enabled privacy-preserving SemCom techniques, including GenAI-assisted semantic-aware friendly jamming and GenAI-assisted semantic covert communication.
• Figure 4: Perspective of agentic AI empowered privacy-preserving SemCom techniques, including agentic reasoning-driven security control, agentic context-aware security design, and agentic evolving security strategy.
• Figure 5: Case studies of GenAI- and agentic-AI-enabled eavesdropping and privacy-preserving SemCom. (a) GenAI-enabled private information reconstruction using GDMs; (b) Agentic RAG-based private information reconstruction; (c) demonstration of behavior and intent inference using agentic AI; (d) GenAI-based semantic-aware friendly jamming; (f) GenAI-assisted semantic covert communication; (e) Agentic context-aware secure SemCom framework