VENENA: A Deceptive Visual Encryption Framework for Wireless Semantic Secrecy

TL;DR

VENENA delivers a practical, end-to-end implementation of physical layer deception for semantic wireless communication by integrating AI-driven visual encryption with image poisoning and power-domain multiplexing. The framework maps confidential messages to semantically meaningful visuals, injects poisoned content to mislead eavesdroppers, and uses a high-power original image alongside a low-power poison mask so that Bob can reconstruct the true message while Eve, especially with full system knowledge, faces degraded or incorrect decoding. Experimental validation on CIFAR-10-based data with a Vision Transformer classifier shows Bob maintaining over $93\%$ message-perception accuracy, while Eve’s success is reduced by more than $40\%$ under informed adversaries. The work analyzes efficiency, scalability, and channel-condition assumptions, and points to future avenues including multi-modal semantic encoding and adaptive channel-aware deception for robust 6G semantic security.

Abstract

Eavesdropping has been a long-standing threat to the security and privacy of wireless communications, since it is difficult to detect and costly to prevent. As networks evolve towards Sixth Generation (6G) and semantic communication becomes increasingly central to next-generation wireless systems, securing semantic information transmission emerges as a critical challenge. While classical physical layer security (PLS) focuses on passive security, the recently proposed concept of physical layer deception (PLD) offers a semantic encryption measure to actively deceive eavesdroppers. Yet the existing studies of PLD have been dominantly information-theoretical and link-level oriented, lacking considerations of system-level design and practical implementation. In this work we propose Visual ENcryption for Eavesdropping NegAtion (VENENA), an artificial intelligence-enabled framework for secure image-based communication. VENENA protects confidential messages by encoding them visually while actively deceiving eavesdroppers: legitimate receivers use artificial intelligence (AI)-based classifiers to extract true message semantics, while interceptors perceive only falsified content. The framework transmits two superimposed image components with different power levels - a high-power decoy image and a low-power correction mask - ensuring only authorized receivers with favorable channel conditions can reconstruct the true message. Experimental validation demonstrates over 93% accuracy for legitimate users while limiting eavesdropper success to 52% even when system design is fully known, validating VENENA's active defense capability for 6G semantic communication.

Figures (3)

• Figure 1: Visual encryption of confidential messages: \ref{['subfig:naive_design']} naive design to defend eavesdroppers without essential knowledge for decryption, and \ref{['subfig:venena']} the venena framework to deceive insider eavesdroppers.
• Figure 2: Sensitivity of message perception accuracy versus mixing ratio $\alpha$ under different transmission power and eavesdropping channel conditions.
• Figure 3: Deception rates of Bob and full-knowledge Eve, respectively, under different mixing ratios in three scenes.