Psychology Congress 2027
Home› Scientific Program
Kadir Uludag
Capital Medical University
Title : The Risk of Subliminal Message Propagation in AI-Generated Videos
Abstract:
The rapid proliferation of AI-generated videos introduces a novel vector for subliminal messaging—visual or auditory cues embedded below conscious perception. Unlike traditional media, generative models can seamlessly insert imperceptible triggers during synthesis, enabling large-scale, personalized, and deniable influence campaigns. This abstract examines the technical feasibility of subliminal injection in diffusion‑based and GAN‑generated videos, the psychological susceptibility to such stimuli, and the resulting ethical and regulatory gaps. We argue that current detection methods are inadequate, and propose a framework for risk assessment, emphasizing the need for transparency standards and real‑time monitoring to mitigate covert manipulation in synthetic media.
Biography:
Dr. Kadir Uludag is a researcher specializing in applied psychology, currently engaged in postdoctoral studies at the Capital Medical University. He earned his Ph.D. in Applied Psychology from the Chinese Academy of Sciences. Dr. Uludağ’s research interests encompass schizophrenia, drug addiction, and educational psychology. He has also explored the intersection of psychology and technology.
4th International Congress on
Psychology & Behavioral
Technical Feasibility: Plausible, but Context-Dependent
The abstract’s technical premise rests on the capacity of generative models to “seamlessly insert imperceptible triggers during synthesis”. This is not merely speculative. The European Commission’s guidelines on prohibited AI practices explicitly recognize that visual subliminal messages in AI systems can involve “images or text flashed briefly during video playback which are technically visible, but flashed too quickly for the conscious mind to register”. The broader literature on deepfakes and synthetic media also notes the use of “subvisual adjustments” and “subaudible distortions” designed to bypass conscious awareness. From a technical standpoint, the injection of such cues into a generative pipeline is conceivable, particularly because diffusion and GAN architectures afford fine-grained control over pixel- and frame-level content.
However, feasibility is not the same as ease of effective deployment. The abstract does not specify the robustness of these triggers under compression, re-encoding, or display variation—real-world conditions that could degrade subliminal signals. Nor does it address whether the influence would be reliably measurable outside controlled laboratory settings. These are empirical questions that the proposed framework would need to operationalize.
