Physical Causality Verification: A Media-Agnostic Instrument for Detecting Synthetic and Manipulated Files

This paper describes a training-free, generator-agnostic instrument for synthetic media detection and deepfake detection. It applies identically to images, audio, and video without modification. Existing deepfake detection and AI-generated media detection tools are empirically grounded - they fail on unseen generators. This instrument does not detect generators. It detects the presence or absence of physical causality, which no generative AI model can fake. The method extracts three physical reads from any file - noise residual (Energy), geometric distortion (Space), and clock timing (Time) - and measures coupling between them using the TwoNN intrinsic dimensionality estimator. Four numbers are produced. The failure modes are analytically derived from rate-distortion theory, not discovered by testing. Novel contribution: the recognition that six existing results across signal processing, information theory, MRI diagnostics, and machine learning were independently describing the same underlying structure - and that assembled together they constitute a complete instrument for media authenticity verification and image authenticity verification. Code available under AGPL at : https://github.com/Amogh-12345/physical-causality-verification