生成式 AI 赋能的安全通信：内生安全到主动防御

The continuous evolution of intelligent and adaptive threats poses severe challenges to the physical layer security of communication systems. Constructing a secure signal processing mechanism with robustness, predictability, and proactivity is an important foundation for achieving trustworthy communications. To promote the development of generative artificial intelligence (GAI)-empowered secure communication technologies, based on a systematic review of existing studies, the theoretical foundations, key mechanisms, and application progress of GAI in secure communications were analyzed and summarized. First, GAI was formalized as a learning-based signal prior for solving inverse problems at the physical layer. Based on this generative prior, endogenous vulnerabilities could be mitigated; physical layer stability was improved, and a reliable security baseline was established through channel prediction, channel state information completion, and hardware distortion correction. Furthermore, a three-layer proactive defense framework composed of intelligent threat perception, dynamic adversarial game, and covert waveform generation based on noise alignment was constructed. Finally, the key challenges and development directions of GAI-empowered secure communications were prospected, including real-time inference latency, the simulation-to-reality performance gap, the introduction of physics-informed constraints, security digital twins, and the construction of autonomous security agents, and the significance of establishing a unified theoretical perspective and an actionable design paradigm for future research was emphasized.