Stanford Internet Observatory releases study on AI-generated CSAM proliferation across clearnet platforms

Researchers at Stanford Internet Observatory published findings on 4 November 2026 documenting the proliferation of photorealistic AI-generated child sexual abuse material across mainstream image-hosting and social platforms, identifying more than 10,000 discrete AI-generated CSAM images uploaded to eight major services during a three-month monitoring period and noting that current hash-based detection systems are largely ineffective because each generated image is unique and does not match known hash databases maintained by NCMEC and other authorities. The study found that offenders are using publicly available and commercial generative AI tools, often with minimal modification, and are sharing both the generated material and the prompts and techniques required to produce it within semi-public communities on Discord, Telegram and Reddit.

The failure of hash-matching to catch novel AI-generated abuse material highlights the need for detection architectures that assess content and context rather than relying solely on known-image fingerprints. Guardii's anti-CSAM module, developed to identify both photographic and AI-generated abuse material, analyses visual indicators and conversational context when images are exchanged in direct messages, flagging high-risk sharing before material is uploaded to public feeds or forwarded to additional victims, and providing escalation to NCMEC and law enforcement where criminal distribution is identified, a capability that addresses the emerging threat documented by Stanford researchers.