Meta Missed 350 AI Child Abuse Ads

It was revealed that Meta had published a large number of ads containing child sexual abuse material that had slipped through its own review system. According to reporting by Wired’s Matt Burgess, based on research by the nonprofit Tech Transparency Project, more than 350 video ads were identified. Some included manipulated images of real children, and minors related to European royalty were among the victims. The ads were placed on Facebook, Instagram, and Threads, having passed through Meta’s ad review process. The company says it reviews all ads in advance, but in this case the system failed to function.

About 50 ads were removed in early August. The company said it had introduced new AI detection measures to strengthen detection and blocking. However, more than 250 additional placements were confirmed even after August, raising doubts about the effectiveness of the measures. Lawmakers have indicated they will launch an investigation. The state of platform ad review and child protection is once again being called into question.

350 Ads That Slipped Through Review

The Tech Transparency Project says that more than 350 abusive video ads placed on Meta’s platforms since late last year have been identified. In the initial investigation, 53 ads were found, and about 50 were removed in early August. Even after that, more than 250 additional placements were confirmed in August alone and beyond. There were also cases where ads identical to those already removed were published again. Each time researchers made a discovery, they reported it to U.S. child protection authorities and through Meta’s reporting channels.

In its own verification, Wired also found that hundreds of explicit undressing ads targeting adults existed on a large scale. These used the same series of promotional tactics. The ads targeting children have already been removed. Meta cited violations of its policies on child exploitation, adult exploitation, and nudity. Nevertheless, the fact remains that they had passed through review at the time of placement.

The ads were delivered in short video format. They were designed to appear while users were browsing. They were actually being displayed at the time researchers reported them. The review system likely relied on post-publication response. The core issue is that pre-publication blocking did not function.

How Images of Real Children Were Misused

What is particularly notable in this investigation is that images of real children were used. In the initial 53 cases, no real children were identified. In ads from August onward, images of multiple real children were confirmed. Official photos of minors belonging to European royalty were repurposed without permission. Photos published on official websites had been manipulated into sexualized videos.

Researchers have not disclosed the name of the royal family to protect the victims. They say they have contacted authorities in the relevant countries. Wired also confirmed that the source of the images was the royal family’s official website. This is not just a problem of AI generation, but also includes misuse using photos of real children as source material. The severity of the harm is at a different level from previous debates.

What is at issue here is not simply AI-generated child sexual abuse material

Katie Paul, director of the Tech Transparency Project, said this. She explained that some of the images are believed to have been obtained from the internet. She criticized the fact that harm spread while advertising revenue was being generated. A structure has emerged in which AI generation and misuse of real photos are mixed. Definitions of AI-generated content are also organized in the AI Glossary: Explaining Opaque recurrence and AGI.

Placements Continued After AI Detection Was

Introduced, With Delays

At the time of the removals in early August, Meta said it had strengthened detection and blocking of harmful ads with new AI measures. It said most of the ads had been placed before the new measures were introduced. However, according to researchers, a large number of similar ads continued to be placed afterward. This means placements continued even after the introduction of AI detection. Mechanisms to prevent re-publication of identical content also did not work sufficiently.

Delays between reporting and removal also came to light. In one case, it took a week to review an ad reported by researchers through Meta’s channels. During that time, the ad continued to be displayed. According to screen recordings, it accumulated hundreds of impressions. The practice of not immediately suspending ads even after reports widened the harm.

Ad review is generally operated through a combination of automated processing and human review. Meta also states in its policy that it reviews all ads. Even so, ads containing child sexual abuse material slipped through. There may have been flaws in the training data for detection models or in the application of review criteria. Improving the accuracy of pre-publication blocking is essential to prevent recurrence.

Undressing Apps Linked to Chinese Developers

Many of the ads were linked to promotion of so-called undressing apps. These are applications that perform processing to make it appear as if clothing has been removed. Researchers pointed to links with multiple apps associated with Chinese developers. The links within the ads were structured to direct users to those apps. Traffic to harmful apps was being generated via Meta’s ads.

Explicit ads targeting adults were also part of the same advertising network. Wired independently confirmed hundreds of related ads. The tactics were similar, combining video with redirection to external sites. Ads targeting children did not exist in isolation. There appears to have been a structural weakness in the review of harmful ads as a whole.

The spread of undressing apps is becoming a problem in many countries. There have been reports of harm from unauthorized manipulation of images of real people. When minors are targeted, it constitutes child sexual abuse material. The fact that the ad delivery infrastructure was used as a channel for abuse is serious. Measures are needed on both the app distribution side and the ad delivery side.

Legislative Investigation and Platform

Accountability

Lawmakers have indicated their intention to investigate in response to this situation. The focus will be on the reality of ad review and the effectiveness of AI detection measures. The fact that Meta earned advertising revenue from the ads will also be a point of contention. Debate over the scope of platform liability is expected to expand. In Europe as well, regulations to strengthen child protection are advancing.

Meta’s advertising business is built on the premise of strict review. Child exploitation is clearly prohibited. Even so, it failed to prevent more than 350 placements. The gap between policy and practice has been exposed. Ensuring transparency is essential to restore trust.

Cooperation between operators is key to identifying senders and preserving evidence. There have been cases where disclosure of information through legal proceedings was used, as shown in Take-Two Subpoenas Discord and Microsoft to Identify GTA VI Leaker. Identifying advertisers and developers will also be a challenge in this case. Sharing reported information with authorities is important. Blocking at the advertiser level is required to prevent re-publication.

Editorial Opinion

On short-term impact: We expect this revelation to prompt a review of Meta’s ad review operations over the next three to six months. Accuracy verification of AI detection and expansion of human review may advance. If a legislative investigation begins, we assess that demands for disclosure regarding advertiser identity verification and prevention of re-publication will intensify.

On the long-term perspective: On a one- to three-year horizon, we expect regulation against the misuse of images of real children to extend to generative AI as a whole. Blocking ad placements and payments for undressing apps is expected to become an industry standard. We believe the introduction of age verification and provenance labeling technologies will become a competitive factor for platforms.

Questions from the editorial team: What caused the failure to prevent more than 350 placements despite claiming pre-publication review? Is an operation that takes a week from report to removal acceptable? To prevent the spread of harm to real children, should the focus be on the advertising infrastructure or on app distribution? We assess that examination of these issues is now required.

References

• “Meta Failed to Catch Hundreds of AI Child Abuse Ads. Some Included Images of Real Kids”, by Matt Burgess — Wired, 2026-09-08T12:00:00.000Z (ARR)
• Source URL: https://www.wired.com/story/meta-failed-to-catch-hundreds-of-ai-child-abuse-ads-some-included-images-of-real-kids/