Stanford Internet Observatory Study: AI-Generated CSAM Increasing Across Platforms Despite Detection Efforts

Research published by the Stanford Internet Observatory identifies a significant increase in AI-generated child sexual abuse material across major platforms, with detection systems originally developed to identify known CSAM proving less effective against synthetic images that do not match existing hash databases. The study documents the use of open-source image-generation models, modified to bypass safety filters, and notes that traditional detection methods relying on PhotoDNA and similar hash-matching technologies cannot identify novel AI-generated material without prior cataloguing.

Hash-matching systems depend on having seen the image before, a model that fails when the material is generated on demand. Detection must instead identify the distribution behaviour and the context in which the material appears. Guardii's anti-CSAM module, designed to recognise both photographic and AI-generated abuse material, monitors the conversational environment in which images are shared, flagging sextortion, grooming and coercion patterns alongside the material itself, and operating across messaging platforms where synthetic CSAM is increasingly produced and traded, so that intervention occurs when the content enters a child's message stream, not months later when a hash is finally added to a database.