Generative AI makes IBSA harder to trace, new forensic study finds

The paper, "Traces of Abuse: How Generative AI Impacts Image-Based Sexual Abuse (IBSA) Investigations," published on arXiv (2608.14616) and accepted at a SOUPS 2026 workshop, systematically analyzes how generative AI changes the forensic landscape of image-based sexual abuse. The authors—Jasmin Wyss, Anna Neumann, Ivy Turk, and Rebekah Overdorf—compare the traces available in four distinct IBSA scenarios, ranging from traditional non-AI abuse to fully AI-generated content. Their core finding: GAI-IBSA produces fewer distinctive digital traces than conventional IBSA, and the traces it does leave are often ambiguous, making attribution and victim identification significantly harder.

The study argues that generative AI benefits abusers in two key ways: it lowers the barrier to creating realistic abusive content at scale, and it breaks the forensic chain that investigators rely on—such as device metadata, editing history, and victim-linked evidence. Because generated images can be synthesized without involving a real victim in the production, standard investigative reasoning from the image itself becomes unreliable. The authors call for updated forensic methods and policy responses tailored to GAI-facilitated abuse. This is one of the first peer-adjacent works to map exactly how GAI disrupts evidence collection, offering a concrete framework for law enforcement and platform moderators.