Meta's AI Detection System Failed to Stop Hundreds of Child Sexual Abuse Ads
Researchers discovered more than 250 ads containing child sexual abuse material on Meta platforms since August, including images of real children, despite the company's new AI filtering tools.
What happened
The Tech Transparency Project identified over 350 abusive video ads running on Facebook, Instagram, Threads, and Messenger since the end of last year, with more than 250 appearing since August. The ads used AI face-swapping technology to morph children's faces into graphic sexual content, then directed users to download "nudification" apps. Researchers identified real-world victims in at least four cases: three from the U.S. (two social media influencers and one stock photo model) and one European royal family member whose official photograph was transformed into explicit material. According to Meta's ad library, the ads were shown to more than 29,000 accounts in the European Union and around 7,000 in the United Kingdom, with more than 250 appearing in the United States, 120 in Australia, and 80 in India where Meta does not publish granular ad performance data. Despite Meta's stated policy of reviewing all ads and deploying new AI tools to detect harmful content, many ads remained live for extended periods—in some cases over a week after researchers reported them—accumulating hundreds of additional views.
Context
Meta's failure to catch this content is significant because it demonstrates that the company's newly deployed AI detection system has not prevented the circulation of child sexual abuse material at scale. The ads were monetized, generating Meta under $5,000 in revenue while potentially harming real minors whose images were weaponized. The discovery reveals a gap between Meta's stated policies against child exploitation and nudification, and the platform's actual enforcement. The involvement of real children—particularly public figures and influencers—rather than solely AI-generated imagery, raises distinct harms related to identity theft, reputation, and victimization. Researchers reported all findings to the National Center for Missing and Exploited Children and are working on victim disclosure.
What's disputed
Meta claims many of the ads had already been removed and most had fewer than 200 impressions, characterizing the findings as limited in scope. The company asserts it does not tolerate nudify apps or child exploitation. However, Meta's response does not address the gap between the timeline of its new AI deployment and the researchers' discovery of hundreds of additional abusive ads, or explain the week-long delays in removing live content after being reported.