Towards Secure Semantic Communications in the Presence of Intelligent Eavesdroppers

TL;DR

This work investigates security challenges in semantic communications for 6G by analyzing intelligent eavesdropping strategies that exploit model inversion and GenAI, under both glass-box and closed-box access to the semantic encoder. It develops a dual defense strategy: (i) GenAI-enhanced eavesdropping analyses to establish strong baseline privacy risks and (ii) a semantic covert communication framework built on an invertible neural network steganography module that hides private semantic content within a host signal. Simulations on facial image data show eavesdropping success rates can exceed 80% and reach 100% with GenAI, while the proposed INN-based covert scheme reduces privacy leakage to 0% and preserves reconstruction quality for legitimate receivers. The results imply a practical path toward secure SemCom in 6G networks by combining adversarial threat modeling with a lightweight, semantically-aware covert defense that maintains QoE for intended users.

Abstract

Semantic communication has emerged as a promising paradigm for enhancing communication efficiency in sixth-generation (6G) networks. However, the broadcast nature of wireless channels makes SemCom systems vulnerable to eavesdropping, which poses a serious threat to data privacy. Therefore, we investigate secure SemCom systems that preserve data privacy in the presence of eavesdroppers. Specifically, we first explore a scenario where eavesdroppers are intelligent and can exploit semantic information to reconstruct the transmitted data based on advanced artificial intelligence (AI) techniques. To counter this, we introduce novel eavesdropping attack strategies that utilize model inversion attacks and generative AI (GenAI) models. These strategies effectively reconstruct transmitted private data processed by the semantic encoder, operating in both glass-box and closed-box settings. Existing defense mechanisms against eavesdropping often cause significant distortions in the data reconstructed by eavesdroppers, potentially arousing their suspicion. To address this, we propose a semantic covert communication approach that leverages an invertible neural network (INN)-based signal steganography module. This module covertly embeds the channel input signal of a private sample into that of a non-sensitive host sample, thereby misleading eavesdroppers. Without access to this module, eavesdroppers can only extract host-related information and remain unaware of the hidden private content. We conduct extensive simulations under various channel conditions in image transmission tasks. Numerical results show that while conventional eavesdropping strategies achieve a success rate of over 80\% in reconstructing private information, the proposed semantic covert communication effectively reduces the eavesdropping success rate to 0.

Figures (9)

• Figure 1: Visualization of the image reconstruction results by the proposed intelligent eavesdropping strategies under Rayleigh eavesdropping channels with an SNR of 5dB.
• Figure 2: Visualization of the proposed intelligent eavesdropping including: 1) Glass-box eavesdropping, where Eve performs MIA to reconstruct the transmitted image using gradient descent. 2) Closed-box eavesdropping, where Eve trains an inverse network using its local dataset to perform MIA. 3) GenAI+Glass-box eavesdropping, where Eve introduces GenAI to produce high-quality candidates and inputs them into the semantic encoder to perform MIA. 4) GenAI+Closed-box eavesdropping, where Eve trains an inverse network to generate latent codes for the GenAI model that can reconstruct the transmitted images with high fidelity.
• Figure 3: Illustration of the proposed semantic covert communication.
• Figure 4: Performance of the proposed eavesdropping strategies under AWGN channels, where SNR varies from 0 dB to 20 dB.
• Figure 5: Performance of the proposed eavesdropping strategies under Rayleigh channels, where SNR varies from 0 dB to 20 dB.

Theorems & Definitions (1)

• Remark 1