Meta's AI Tools Miss Hundreds of Child-Abuse Ads on Its Platforms

In early last month, Meta removed roughly 50 advertisements that directly featured child sexual-abuse material (CSAM) from Facebook, Instagram and Threads. Researchers at the nonprofit Tech Transparency Project (TTP) later reported that more than 250 additional ads containing CSAM have run on Meta’s platforms since the start of August. Unlike the 53 ads initially uncovered, the newer ads include images of real children, some of which were identical to the material Meta had already taken down. One particularly disturbing example repurposes an official photograph of a minor from a European royal family into a video depicting a graphic sexual act.

Wired Magazine

Meta had recently rolled out new artificial-intelligence tools intended to “better detect and block” harmful ads. The company explained that the majority of the abusive ads discovered earlier were posted before these AI systems were deployed, implying that the tools should have prevented later occurrences. Researchers, however, say the AI has not been fully effective, as the volume of CSAM ads has not declined and new instances continue to appear.

Wired Magazine

• Approximately 50 CSAM ads deleted in early last month.

Wired Magazine

• 53 ads initially identified by TTP researchers.

Wired Magazine

• Over 250 additional CSAM ads detected since early August, bringing the total to more than 300 offending ads.

Wired Magazine

• The newer ads contain verifiable images of real children, including a royal minor whose likeness was altered for sexual exploitation.

Wired Magazine

Meta asserted that its AI detection system was launched after most of the previously identified abusive ads were posted, suggesting that future ads should be caught automatically. The company has not provided a detailed breakdown of the system’s performance metrics. TTP’s director, Katie Paul, emphasized that the problem extends beyond AI-generated content, noting that many of the offending images appear to have been harvested from existing web sources before being embedded in the ads. TTP officials have contacted authorities in the royal family’s country to address the misuse of the official photograph.

Wired Magazine

• “What we’re talking about here is not simply AI-generated child sex abuse material,” — Katie Paul, the director of the TTP

Wired Magazine