Flawed by Design: The Leinster Principle and the Forensic Challenge of Synthetic Media

As the complexity of synthetically generated media continues to increase, it’s often assumed that sufficiently advanced systems will eventually produce content that is perfect and therefore undetectable. However, by its very perception of being perfect, AI-generated media may appear anomalous when compared with real-world media. We are already in a liminal zone where anything can be denied as fake, or dismissed as synthetic, regardless of whether it is or is not [1]. As humans we believe what we see quicker than what we hear. We tend to believe something is ‘fake’ when it is out of context: the classic six fingers on one hand comes to mind. That’s the purpose of the Leinster Principle, where the imperfections of the synthetic media mean it is harder to detect when in context. Drawing inspiration from the architectural false facades of Leinster Square Gardens in London, this study evaluates whether the introduction of such imperfections can influence automated detection systems, whether it be in the correct or indeed incorrect direction. After applying several common signal processing methods, the results indicate that these disguises can substantially influence detector behaviours, producing changes in classification outcomes of up to 100%, with ambient noise producing an absolute mean change of approximately 35%. These findings provide preliminary empirical support for the proposed principle and suggest that contextual imperfection may present an effective adversarial strategy against synthetic media detection systems.