Meta's AI Detection Systems Fail as 332 Child Abuse Ads Slip Through Facebook and Instagram

Meta Fails to Stop Hundreds of AI Child Abuse Ads Despite New Detection Tools

Meta has failed to prevent over 332 AI-generated child sexual abuse ads from running across Facebook, Instagram, and Threads since late last year, according to new findings from the Tech Transparency Project1

1

. The discovery comes weeks after Meta deleted approximately 50 similar ads and claimed most had appeared before the company rolled out new AI tools to block harmful content1

1

.

The scale of the problem reveals a critical weakness in Meta's content moderation systems. More than 250 additional ads containing CSAM have run across Meta's platforms since August alone1

1

. Unlike previous discoveries, researchers identified images of real children in these new ads, marking a disturbing escalation in how AI child abuse material is being created and distributed.

Real Children Identified in AI-Generated Sexual Abuse Imagery

Tech Transparency Project researchers traced the manipulated images back to four real minors whose ordinary photographs were weaponized by AI face-swapping apps1

1

2

2

. Three victims are from the United States, including two teen influencers with public social media accounts and one preteen whose image appeared on stock photo websites1

1

. The fourth victim is a member of a European royal family, whose official photograph was transformed into explicit video content1

1

.

"What we're talking about here is not simply AI-generated child sex abuse material," says Katie Paul, director of the Tech Transparency Project1

1

. Paul notes that images appear to have been taken from the web before being used in the ads, with Meta profiting throughout the process.

Generative AI image tools can modify existing photographs while preserving recognizable features, meaning no explicit original photo is required2

2

. Once an ordinary image becomes publicly accessible, someone can download it and feed it into AI tools designed to alter photographs into sexual abuse imagery.

How the Ads Evaded Meta's Detection Systems

The ads followed consistent patterns across Facebook, Instagram, Messenger, Threads, and Meta's broader advertising network. Most featured innocuous pictures of preteen or teenage girls with overlaid text claiming "not just a photo" and "there are no restriction" on what could be done with the image1

1

. When the video ads played, children's faces were morphed into graphic sexual videos using AI image tools. The ads directed users to download nudification apps from Apple and Google's app stores, with many linked to Chinese developers1

1

.

Meta claims all ads undergo review before running and that automated systems play a major role in the process2

2

. However, researchers documented instances where it took Meta a week to review live ads reported through its own tools1

1

. During that time, ads accumulated hundreds of additional views. In one case involving a real teenager, an ad had reached four European accounts when reported to Meta but had reached over 330 accounts by the time Meta removed it days later2

2

.

Meta spokesperson Tracy Clayton told WIRED that bad actors deliberately attempt to evade detection systems and some videos were harder to detect because images of children appeared only briefly or were blurred2

2

. The company said many ads had already been removed, most received fewer than 200 impressions, and it received less than $5,000 in payments from the ads2

2

.

Widespread Reach and Policy Violations Across Meta Platforms

The ads collectively reached more than 29,000 accounts across European Union countries and approximately 7,000 additional accounts elsewhere1

1

. All violated Meta's policies on child exploitation, adult exploitation, and nudity, yet were initially published despite the company's claim that it reviews all ads1

1

.

When some ads containing child sexual abuse were removed, Tech Transparency Project says they did not initially include disclosures indicating policy violations related to child sexual abuse. Those details only appeared after WIRED questioned Meta about the discrepancy1

1

.

Researchers reported all ads to NCMEC's CyberTipline, which allows reporting of online child sexual abuse material1

1

. They are consulting on how to appropriately disclose their discoveries to those impacted. WIRED independently identified hundreds of other related nudification ads containing explicit content of adults on Meta's platforms1

1

.

Mainstream Apps Facilitate the Threat to Protect Children Online

Researchers found ads directing people toward nearly 50 apps available through Apple's App Store and Google's Play Store2

2

. Around 300 of the roughly 350 ads examined promoted AI image or video generators2

2

. Apple told WIRED that some developers initially submitted compliant app versions before subsequently adding prohibited nudification capabilities2

2

. Apple said it removed 17 of 20 apps identified by researchers and took action against the remaining three, plus seven additional apps2

2

.

"AI apps are making it far too easy to create realistic criminal imagery of child sexual abuse, creating the new risk of everyday photos of children being taken and transformed into sexual content," Emma Hardy, director of communications at the Internet Watch Foundation, told WIRED2

2

. The findings underscore how generative AI has fundamentally changed the risks around sharing photos of children online, as creating convincing fake imagery no longer requires explicit photographs.