Privacy Preserving in Wiretap Channels Beyond Shannon’s Paradigm: Current Trends and Future Challenges

Shannon’s classical information theory has long underpinned communication systems by ensuring bit-level fidelity, yet it remains agnostic to the semantic content of transmitted data. Semantic Communication (SemCom) introduces a paradigm shift beyond Shannon’s model by enabling the transmission of meaning-relevant information, thereby enhancing communication efficiency. By leveraging Artificial Intelligence (AI), SemCom prioritizes task-relevant features, achieving significant reductions in bandwidth usage and latency, while improving throughput and system performance. Despite these advantages, SemCom introduces novel security vulnerabilities. The inherent openness of wireless channels, high correlation between source and transmitted representations, and structural weaknesses in AI models render SemCom particularly susceptible to eavesdropping and privacy breaches. These risks necessitate a reassessment of traditional security paradigms. This paper presents a comprehensive survey of emerging security strategies for SemCom against eavesdropping, encompassing physical-layer security, cryptographic frameworks, data-driven approaches, and semantic-level covertness. Each class of techniques is analyzed with respect to its methodological foundations, effectiveness, limitations, and its deviation from conventional communication security mechanisms. Distinctively, this survey provides a comparative analysis of these strategies under model inversion-based eavesdropping attacks, offering a unique perspective absent in existing literature. The paper concludes by identifying critical research challenges and outlining prospective directions for secure and trustworthy SemCom systems.