Abstract: The heightened verisimilitude of Artificial Intelligence (AI) technology is a double-edged sword, simultaneously empowering societal progress and harboring risks that undermine the authenticity of information. This paper focuses on the cutting-edge issue of Deceptive AI, systematically constructing an innovative classification framework based on the mechanism of technical verisimilitude and risk dimensions to address the increasingly severe challenge to authenticity. Transcending fragmented and isolated studies of phenomena such as deepfakes and social bots, this paper pioneeringly defines the concept of “Deceptive AI” within the Chinese context. It profoundly distinguishes and elucidates the differences and connections between “Deceptive AI” and “deception by AI,” underscoring that its core lies in the technical verisimilitude of AI in simulating intelligent behavior and the induced cognitive distortion effect. This research dissects the conceptual connotation and intrinsic mechanisms of technical verisimilitude, and categorizes five types of deceptive AI technologies—Generative Adversarial, Behavior Simulation, Semantic Manipulation, Cognitive Intervention, and System Contamination—revealing their spectrum-like evolutionary features. The study constructs a two-dimensional, four-quadrant classification framework based on “Technical Verisimilitude - Deceptive Intent,” delineating the deception mechanisms and differentiated risk spectra of Fraudulent, Simulation, Functional, and Inducing AI. High-verisimilitude AI is not merely a technical tool, but rather deeply embeds itself within the human cognitive system. It triggers profound psychological mechanisms such as cognitive shortcuts, emotional resonance, echo chambers, and authority bias, inducing collective cognitive distortion and consequently posing systemic threats to the information ecology, social trust, and even public safety. This research not only provides an original theoretical framework for understanding and addressing the authenticity risks of deceptive AI, but also aims to alert society to the cognitive risks inherent in technical verisimilitude, offering forward-looking perspectives and risk response strategies for the healthy development of a future era of human-AI coexistence.