Security and Privacy in Edge-Based Semantic Communication: Mechanisms, Challenges, and Future Directions

Edge computing and semantic communication (SC) are two key technologies for future communication systems. Edge computing reduces transmission delay and the amount of data sent through the network. SC improves communication efficiency by extracting and transmitting task-relevant semantic information rather than raw bit streams. Their integration is expected to support low-latency and bandwidth-efficient services in 6G networks. However, edge-based SC also raises new security and privacy concerns. It relies on deep learning models, distributed edge data, and compact semantic representations that may contain high-value task information. As a result, attacks may target not only wireless links but also semantic data, learning models, and edge resources, especially during collaborative model training. This article reviews representative security and privacy mechanisms for edge-based SC from two practical stages: semantic model training and semantic information transmission. For model training, we focus on threats such as poisoning and Sybil attacks in federated learning-based collaboration among edge devices and edge nodes. For information transmission, we discuss semantic-level privacy protection and physical layer security as complementary defense mechanisms. We further compare existing mechanisms in terms of targeted threats, impact on semantic recovery accuracy, communication overhead, and suitability for resource-constrained edge devices. Finally, we outline open challenges and future research directions, including lightweight security design, the trade-off between privacy protection and semantic utility, multi-layer collaborative defense, and digital twin-assisted resource management.