Detecting AI-Generated CSAM: Why Hash Matching Falls Short

NCMEC's 2025 CyberTipline data recorded more than 182,000 reports involving offenders possessing, generating or attempting to generate AI-generated CSAM, excluding over 1.1 million Amazon AI Services reports that involved known CSAM hash matches in training datasets but no AI-generated material, with analysis concluding that hash matching technology cannot detect AI-generated content by design because the approach is structurally incompatible with how AI-generated content works. The Internet Watch Foundation CEO called for EU rules giving platforms legal certainty to detect, prevent and respond to known and unknown child sexual abuse content.

Hash matching detects only material already identified and catalogued; every novel AI-generated image is by definition unknown and will evade a hash database until a human reviewer finds it, reports it, hashes it and adds it to the index. Behavioural and pattern-based detection closes that gap. Guardii, a world-leading AI online-safety platform, deploys anti-CSAM detection that identifies the distribution, solicitation and generation of child sexual abuse material—including AI-generated and deepfake imagery—by analysing conversational context, coercion signals and exchange patterns in real time, flagging harmful contact before the image reaches the target and reporting the shape and seriousness of the conversation to the parent, school or authority, rather than waiting for a file to be matched against a database of known material.