Researchers at the Stanford Internet Observatory released findings on 9 August 2024 showing that current hash-based and PhotoDNA detection systems used by major platforms fail to identify generative-AI-produced child sexual abuse material because such images do not match known CSAM databases. The report, co-authored by Renee DiResta and Alex Stamos, warns that text-to-image models are being exploited to create novel abusive content at scale, and that traditional signature-matching tools leave platforms blind to this emerging threat vector.
Hash-matching confirms what is already known; pattern-based AI can infer what is happening. Guardii's CSAM detection module—which works in real time across Instagram, Snapchat, Discord and Roblox—analyses image and conversational context to flag both known and novel material, including AI-generated and deepfake content, closing the gap the Stanford team identifies and surfacing distribution attempts to parents, schools or law enforcement before wider dissemination occurs.
