AI Child Abuse Ads Evade Meta Detection, Some Featuring Real Children

Researchers identified over 350 advertisements on Meta’s platforms that directed users to approximately 50 apps on Google Play and Apple App Store. Around 300 of these ads promoted AI image or video generation tools, with most linked to developers based in China. Several apps had already been removed by the time the investigation was reported.

Among the problematic apps were eight identical applications on Apple’s App Store associated with Chinese developers. These apps generated explicit AI content, including a menu of sexual scenarios such as one labeled “threatening knife,” which depicted violence against women. WIRED also discovered seven additional apps with identical designs, explicit content, and developer details.

Apple spokesperson Adam Dema stated that the company has zero tolerance for developers evading review processes. The developers initially submitted compliant apps but later introduced unauthorized “nudify” features. Apple removed 17 of the 20 apps mentioned in the research and took action against the remaining three, along with seven apps found by WIRED.

Google spokesperson Dan Jackson confirmed the company prohibits apps with sexual content or nudification capabilities. Google removed all apps in the report, suspended hundreds of violating apps, and restricted search terms like “nudify” as part of its proactive enforcement efforts.

David Pham, a former Meta employee who investigated nonconsensual imagery, noted a year-over-year increase in nudify apps from Chinese developers. He observed similar ads during his time at Meta, with uploaders constantly adapting to bypass safety systems.

Pham described how uploaders quickly circumvented new detection methods—often within days—by changing domains, obscuring identities, or altering images. Despite Meta’s resources, he emphasized the need to prevent CSAM and NCII material from reaching users’ devices.

Meta’s ad moderation systems appear insufficient, with over 11,000 undress ads appearing since February, many from repeat offenders. WIRED’s analysis found ads using similar language to those in TTP research. Meta acknowledged adversarial tactics and stated it has incorporated the identified language into its moderation systems.

An ad featuring a teenager whose identity was known was reported after reaching four European accounts but took days to remove, ultimately seen by over 330 users. Paul from TTP criticized the delay, stressing that such issues should not require nonprofit intervention given the company’s vast resources.

Source link