Generative models have reached a point where they can synthesise video that is hard to distinguish from genuine footage. This poses a serious challenge for forensic authentication, as traditional methods may not keep pace with the rapid evolution of AI generation techniques.
Learned detectors, which typically key on artefacts left by a specific generator, degrade sharply when confronted with unseen models. This limitation makes them unreliable in real-world scenarios where the exact generative model is often unknown.
The paper introduces an agentic forensic framework built around Electric Network Frequency (ENF) signals, which are naturally embedded in many video recordings. By analysing these signals, the framework aims to authenticate video in a way that is less dependent on generator-specific artefacts, potentially offering a more robust defence against AI-generated forgeries.
The abstract is brief, but the work signals a shift from artefact-based detection toward physical-signal-based authentication. Further technical details are not provided in the available summary.