Meta's Facebook and Instagram platforms hosted more than 300 paid advertisements featuring child sexual abuse material (CSAM) in 2024, including AI-manipulated images of real children, according to an investigation by the Tech Transparency Project (TTP). The findings have intensified scrutiny of Meta's advertising safeguards and its ability to detect AI-generated child exploitation content at scale.

What the Investigation Found

According to TTP, investigators identified photographs of real children — sourced from social media posts and public websites — that had been processed through generative AI tools to fabricate explicit sexual imagery. Most of the advertisements followed a disturbing pattern: they opened with a clip of adult pornography before transitioning to an AI-manipulated image of a child.

In one particularly egregious case flagged by the investigation, a publicly released photograph of a minor member of a European royal family was digitally transformed into a video depicting the child in a graphic sex act. That advertisement promoted a Chinese AI application and was reportedly targeted at adult male users on both Facebook and Instagram. The children depicted ranged from toddlers and preteens to young teenagers, the investigation noted.

Links to Chinese Advertising Partners

The investigation further found that several of the advertisements had been placed by Meta's advertising partners based in China, including at least one Chinese state-controlled company. The vast majority of the ads directed users toward AI applications developed in China, with many of those apps offering or enabling so-called 'nudification' capabilities — tools that digitally remove clothing from images or generate sexually explicit versions of photographs.

Notably, the recurring use of similar captions, voiceovers, and background music across multiple advertisements suggested a coordinated or templated distribution effort, raising questions about whether the content was being systematically produced and distributed rather than uploaded by isolated bad actors.

Meta's Response and Policy Gaps

Meta did not initially respond to TTP's request for comment. After the organisation shared its findings directly with the company, Meta conducted a rapid cleanup of its advertising ecosystem. Approximately 150 advertisements that were still accessible in Meta's Ad Library were removed, according to TTP.

Meta also updated records for more than 100 advertisements that had already been taken down, reclassifying them as violations of its child-exploitation policies. Critically, those advertisements had previously been categorised under Meta's adult-oriented content policies — without any explicit identification of children's involvement — raising concerns that Meta's enforcement systems may be systematically undercounting child sexual exploitation cases, particularly where AI-generated or AI-manipulated imagery is involved.

A Meta spokesperson said the company does not tolerate 'nudify apps or any kind of child exploitation, whether real or AI-generated,' adding that many of the advertisements flagged by the investigation had already been removed and that the company continues to strengthen its detection and enforcement systems.

Broader Implications for Platform Safety

The TTP investigation arrives at a moment of heightened regulatory pressure on major platforms over child safety. This comes amid ongoing legislative debates in multiple jurisdictions — including the United States and the European Union — over platform liability for AI-generated harmful content. Critics argue that Meta's reactive cleanup, rather than proactive detection, underscores a structural failure in how large platforms police paid advertising, where financial incentives can inadvertently subsidise harmful content distribution.

The case also highlights a growing vulnerability: as generative AI tools become more accessible, the barrier to producing synthetic CSAM is falling, while detection systems at major platforms have not kept pace. Whether Meta's updated enforcement classifications will translate into a meaningful reduction in such content — or merely a reclassification exercise — remains to be seen.