Meta let roughly 350 AI-generated child sexual abuse ads slip through its review process, and some were built from photos of real children.
At least some of the abuse images in those ads used real children's faces rather than fully synthetic ones. One of the children pictured is a member of a European royal family, showing how easily identifiable real kids can end up inside generated abuse content. Lawmakers say they now plan to investigate how the ads made it past Meta's ad-review system undetected. It isn't clear from what's been reported how many of the 350 ads relied on real photos versus entirely synthetic images.
The deeper issue, as we see it, isn't a one-off miss - it's that ad-review systems built to catch known, previously cataloged abuse imagery are now facing generative tools that can produce new material instantly, sometimes assembled from real photos scraped elsewhere online. That's a much harder pattern for automated filters to fingerprint than reused CSAM, and it raises the question of whether moderation tooling built for one kind of abuse content is equipped for images invented fresh for every ad.
If a platform with Meta's moderation resources let hundreds of these through, smaller ad networks with far less to spend on review are likely missing more.