Meta Ran Ads Featuring AI-Generated Child Sexual Abuse Material

Citing data from the research group Tech Transparency Project (TTP), WIRED reports that over the past nine months Meta cleared for display dozens of paid ads containing child sexual abuse material, including AI-generated images. The ads ran on Facebook, Instagram, Messenger, and Threads and were targeted at users in the United States, the United Kingdom, and more than a dozen European countries.

In Meta's ad library — a transparency tool that catalogs the ads that have been shown — researchers found over 50 images and videos with impermissible content. Some of them led to so-called nudify apps, which “undress” people in photos. As TTP director Katie Paul stresses, this is not user-generated content but specifically advertising that Meta reviewed, approved, and cleared for display, getting paid for it in the process.

How it worked

According to the researchers, one of the videos used footage of a child alongside a caption offering to “bring fantasies to life” with the help of AI; when tapped, the child's face was inserted into a scene depicting a sexual act. Another ad was published repeatedly with an image of a girl in an explicit pose. Most of the clips were no longer actively running, but they remained available in the library for months. At least one ad reached more than 2,500 accounts in Europe, and the company does not disclose the full figures for the United States.

Response and scale

After WIRED reached out, the company removed the ads as violations of its rules. A Meta spokesperson said that the reach of most of them was minimal, that many had been disabled in advance, and that some predated recently deployed AI tools for blocking violations at upload. The company also noted that over the past year it had removed more than 36 million pieces of child sexual exploitation content and had filed lawsuits against the developers of nudify apps.

The new tools, however, did not work completely: while the story was being prepared, TTP found around 30 more ads, many of which appeared after WIRED's inquiry and were active. Paul notes that among them were previously removed clips — meaning Meta's systems are capable of recognizing identical ads but do not always take them down.

Most of the ads were placed by accounts with almost no followers; where an advertiser was listed, the trail led to Chinese companies. Some of the clips advertised the App Store app MaskAI, which Apple removed after journalists reached out. Earlier, in July, the BBC had found ads on Instagram offering such material for sale in India.

Alexios Mantzarlis, an expert who has reported on more than 25,000 nudify-service ads on Meta's platforms, attributes the resilience of this industry to the platforms' “timid enforcement”: the perpetrators quickly create new accounts and domains, operating along the lines of fraud networks.

Source: Hacker News