Meta Said Its AI Would Catch Child Abuse Ads. Researchers Found 250 More, Some With Real Kids.

The 53 AI-generated child sexual abuse ads that Meta removed from Facebook, Instagram, and Threads in early August were a partial count. The Tech Transparency Project has since identified more than 250 additional such ads on Meta's platforms, including advertisements that depict an identifiable member of a European royal family whose official portrait was transformed into a video of a graphic sex act and then promoted as a paid ad.

WIRED independently confirmed the source image used in the royal-family ad was pulled from the official website of the family. The minor's identity is being withheld by TTP and WIRED to protect the victim. The point is not who the child is; it is that an officially released photograph of any child can be turned into abuse material and run, with payment, on the world's largest social advertising system.

TTP director Katie Paul framed the case in plain terms: "What we're talking about here is not simply AI-generated child sex abuse material." The researchers are reporting the ads to US child safety authorities and through Meta's reporting tools each time they are found live. WIRED's tally of the post-August period now puts the total number of CSAM ads on Meta properties since the end of last year above 350.

Meta's response has leaned on a sequencing argument: most of the deleted ads, the company said, predated the early-August rollout of its new AI-detection tooling. The post-August count of 250+ directly undercuts that claim. The new tools were live while the documented ads were running, and the ads researchers are still flagging share one shape: a real minor's photograph, an AI-generated transformation into a sex act, and a paid promotion through Meta's ad system.

The new tooling is a content classifier. It screens the rendered video. The threat researchers are now documenting starts with the source material underneath. A royal family press photo, a school portrait, or any photograph a parent has uploaded to a public website can become the substrate of a paid abuse ad. A classifier that screens the finished video still sees a moving image of a minor; the question of whether that minor exists, or whether the body in the frame matches a real child whose image was scraped from a public source, is a different lookup the current tooling does not perform.

Many of the ads link to "nudification" apps tied to Chinese developers, according to TTP. The supply chain is consistent: a real image goes in, an AI video comes out, the ad is paid for in US dollars on Meta's system, and the click-through lands on an app store or install page that monetizes the underlying act. Meta's ad review and content review do not share a system. The ads run on the seam between them.

The regulatory backdrop is tightening. Sen. Mark Warner has previously pressed Meta's CEO on CSAM and non-consensual intimate imagery ads; a New Mexico court final judgment in case D-101-CV-2023-02838 has put the company under a separate state-court order that the Tech Oversight Project summarized in an August fact sheet. Meta's own H1 2026 and Q4 2024 Integrity Reports describe the policy posture the company claims to operate under; the post-August 250+ count is now the empirical test of how that posture performs in production.

How many of the 250+ ads were active after Meta's early-August detection rollout, and how many were taken down only after TTP or WIRED flagged them by hand. That ratio is the public record of whether Meta's new tooling closes the loop on real-minor source material, or only on the AI-output layer that was already the easier problem.

Editor's note: WIRED's reporting contains descriptions of child sexual abuse imagery; this analysis mirrors that sensitivity and does not reproduce or describe the imagery itself.