ScholarWorks@동국대학교

초록

Semantic communication (SemCom) has emerged as a task-oriented communicationparadigm that prioritizes meaning delivery over exact bit recovery. The integration ofgenerative artificial intelligence (GenAI) into SemCom further enables knowledge-guidedinference, multimodal reconstruction, and semantic compression through architecturessuch as large language models, variational autoencoders, generative adversarial networks,and diffusion models. At the same time, this integration introduces new security and pri-vacy risks, including semantic eavesdropping, model inversion, semantic jamming, covertbackdoors, prompt manipulation, and knowledge-base leakage, which are not adequatelycaptured by conventional communication security models. In this survey, we provide asecurity-centric review of GenAI-assisted semantic communication systems by organizingthe literature according to threat models, attack surfaces, defence strategies, and semanticmodalities across text, image, and multimodal settings. The survey was conducted usingIEEE Xplore, ACM Digital Library, SpringerLink, arXiv, and Google Scholar. Approxi-mately 180 papers were initially screened, and 53 representative studies published between2021 and 2026 were selected for detailed review. Based on this analysis, we classify themajor threats into adversarial perturbation, jamming, poisoning and backdoor attacks,privacy leakage and semantic eavesdropping, and generative-model-specific vulnerabili-ties involving diffusion, large language models, and multimodal foundation models. Wefurther map the corresponding defences, including adversarial training, model ensembling,semantic-aware encryption, diffusion-guided denoising, privacy-preserving representationlearning, and secure resource allocation. The survey also identifies persistent open chal-lenges, including the lack of standardized semantic security metrics, unified benchmarks,cross-layer evaluation frameworks, and robust defences for GenAI-native and multimodalsemantic communication systems. Overall, this work provides a structured reference forthe design of secure, trustworthy, and attack-resilient generative semantic communicationsystems for future intelligent networks.

키워드