Meta ran hundreds of child abuse ads after launching AI detection

Watchdog researchers identified over 300 paid video promotions depicting minors that reached tens of thousands of platform users across multiple continents.

Automated defenses bypassed by altered images

Meta allowed more than 300 paid advertisements depicting child sexual abuse material to run across its platforms this year, according to findings published by the Tech Transparency Project.¹² The watchdog group identified the material in Meta's public ad library after the social media company claimed new artificial intelligence screening tools would catch and block prohibited uploads.¹²

The discovery follows an earlier report in August that identified 53 similar advertisements across Facebook, Instagram, Messenger, and Threads.¹² At the time, Meta removed the flagged promotions and stated that most of the abusive campaigns predated the rollout of its updated detection models.¹³ Tech Transparency Project researchers subsequently documented more than 250 additional advertisements containing explicit scenes involving minors that ran on Meta platforms after that announcement.¹²

Unlike the initial batch of computer-generated images, researchers confirmed that several of the newer advertisements incorporated authentic photographs of real children.¹² In one instance, an official portrait of an underaged member of a European royal family, released publicly the previous year, was animated into a video depicting a graphic sex act.¹² Tech Transparency Project did not name the royal to protect the minor and notified authorities in the country, while outside verification confirmed the source image originated on an official royal website.¹

Researchers identified three other instances where authentic images of living minors were manipulated.¹ Two cases involved teenage influencers with public social media accounts on Meta platforms, including a 14-year-old girl who had shared a photograph wearing a sports uniform.¹² A fourth case used a commercial stock photograph of a preteen model wearing an athletic outfit with pigtails, which the advertisement converted into video depicting sexual abuse by an adult male.¹²

Network mechanics and app distribution

The advertisements followed consistent visual templates designed to promote face-swapping software. Campaigns opened with ordinary photographs of preteens or teenagers, accompanied by captions stating that the image was not just a photo and that no restrictions applied to what could be generated.¹² When users played the short video clips, the children's faces were digitally mapped onto explicit adult footage or animated into sexual acts.¹²

The paid placements directed users to download generative software listed on Apple's App Store and the Google Play Store.¹² Of the 332 advertisements identified by researchers, 298 linked directly to image or video generation applications, with 253 of those tools originating from developers based in China.²¹ Many of the listings functioned as nudification software, designed to strip clothing from uploaded portraits or place individuals into nonconsensual pornographic clips.²¹

Several applications operated across identical codebases while using different developer names. Tech Transparency Project found eight matching applications on the App Store that shared templates and user interfaces, including tools marketed under names such as Lumin, Craftor, Voyag, and Mask AI.²¹ When opened, the software presented users with menus of explicit scenarios, including depictions of violence against women.¹²

Apple removed 17 of the 20 applications cited in the research after receiving inquiries, and later took action against remaining listings.¹ Google stated that it suspended hundreds of violating applications and restricted search terms like nudify across the Play Store. Spokespersons for both companies maintained that developer guidelines prohibit sexually explicit material and unauthorized manipulation tools.

Reseller pipelines and moderation delays

A significant portion of the abusive advertising entered Meta's network through authorized intermediary agencies in China.²¹ Because Meta platforms are blocked in mainland China, the company relies on certified reseller partners to manage Chinese advertising budgets, paying agencies a 10 percent commission on placed campaigns. Resellers channel billions of dollars into Meta advertising annually.²

Tech Transparency Project discovered that three authorized resellers placed advertisements containing identical child sexual abuse material.² These included Guangdong Advertising Group, a state-controlled entity known as GIMC that Meta previously listed as an agency success story, as well as Meetsocial and BlueFocus.² One placement handled by GIMC promoted a virtual private network called AstralVPN on Google Play, which had recorded over 100,000 downloads.

The reseller advertisements used empty shell accounts on Facebook that displayed generic profiles and minimal followers, avoiding standard platform checks for inauthentic behavior.² In internal reporting documented by Reuters, Meta exempted authorized resellers from immediate enforcement actions and whitelisted their purchases, which permitted flagged campaigns to remain active pending secondary reviews.²

Moderation response times allowed live campaigns to accumulate audiences after being reported.¹ Tech Transparency Project submitted immediate reports to Meta and to the CyberTipline operated by the National Center for Missing and Exploited Children.¹² In several instances, Meta took up to a full week to remove active promotions, during which individual advertisements reached hundreds of additional users.¹

Audience reach and policy reclassification

Meta's regional transparency archives recorded significant distribution across European jurisdictions. The advertisements reached more than 29,000 accounts across the European Union and approximately 7,000 accounts in the United Kingdom.¹² Because Meta removes ad data for non-political campaigns in the United States once a promotion concludes, comprehensive domestic audience metrics remain unpublished.²¹

Geographic targeting parameters showed that 262 of the 332 identified campaigns targeted users in the United States.² Another 120 campaigns appeared in Australia, and 84 ran in India.²¹ The Indian placements appeared despite prior directives from the Indian government instructing Meta to purge all commercial promotions linked to child exploitation material on Instagram.²

Meta removed roughly 150 active listings from its public library shortly after receiving the watchdog's compilation on September 2.² Following the removals, the company revised 113 historical violation records.² Those notices previously cited adult nudity or sexual solicitation rules without recording the involvement of minors, updating the classification to reflect child sexual exploitation only after researchers pointed out the omission.¹

Former Meta abuse investigator David Pham noted that commercial creators of illicit tools continuously modify technical assets to evade automated filters.¹ Developers alter domains, obscure account registration data, and introduce minor pixel distortions or video blurs to prevent algorithmic fingerprinting.¹ Pham observed that operational teams repeatedly adapted within five days of new detection rules being applied across platform systems.¹

Regulatory scrutiny and legal exposure

Meta defended its enforcement record by stating that the abusive campaigns represented a small fraction of overall network traffic. Meta spokesperson Tracy Clayton said the company does not tolerate child exploitation or nudification tools, adding that most identified advertisements received fewer than 200 impressions and generated under $5,000 in total revenue. Clayton stated that the brief appearance of minors and the inclusion of adult footage complicated automated detection.¹

The findings attracted immediate attention from state and federal regulators. Virginia Senator Mark Warner sent a formal letter to Meta chief executive Mark Zuckerberg demanding an explanation for why commercial screening systems failed to intercept illegal media.¹ Attorneys general in Michigan and Florida opened formal inquiries into Meta's advertising review procedures, and Australia's eSafety Commissioner requested senior briefings regarding compliance with platform safety standards.¹

The revelations coincide with existing legal challenges concerning child safety on Meta properties. Meta previously agreed to pay up to $16.7 billion to settle claims brought by state attorneys general regarding platform harms to minors, and the company is appealing a New Mexico court ruling that mandates human moderation staff and upgraded reporting systems within the state.¹ While federal statutory immunities generally protect web platforms from civil claims regarding third-party posts, federal criminal statutes strictly prohibit the distribution, advertisement, and monetization of child sexual abuse material.²

Reporting note: this piece draws on reporting by Matt Burgess for Wired published September 8, 2026, and investigative findings released by the Tech Transparency Project. Meta audience reach figures and platform records are derived from the Meta Ad Library.

Source: Wired via Condé Nast, September 8, 2026.

References

This article is based on 7 sources, listed in the order they are cited.

1. 1 MB Matt Burgess announcement · 8 Sep 2026 Meta Failed to Catch Hundreds of AI Child Abuse Ads. Some Included Images of Real Kids See the source
2. 2 T techtransparencyproject.org TTP - Meta Ran Hundreds of Paid Ads with Child Sexual Abuse Imagery See the source
3. 3 W wired.com Meta Ran Ads That Contained AI-Generated Child Sexual Abuse Imagery See the source
4. 4 T techtransparencyproject.org Tech Transparency Project See the source
5. 5 T techtransparencyproject.org TTP - Facebook See the source
6. 6 X x.com Tech Transparency Project (@TTP_updates) on X See the source
7. 7 X x.com Tech Transparency Project (@TTP_updates) on X See the source