Meta took days to remove advertisements containing AI-generated child sexual abuse material (CSAM) on Facebook and Instagram. Some of these ads used photos of real children, including a press photo of a young member of a European royal family and images taken from a popular Instagram profile of a preteen girl.
The investigation, published Tuesday by the Tech Transparency Project (TTP), found that Meta failed to detect 332 ads containing CSAM this year. The "vast majority" of these ads promoted AI apps made in China, while many others pushed so-called "nudify" apps.
What Are Nudify Apps and How Do They Work
Nudify apps use artificial intelligence to digitally alter images of people. In this case, bad actors used these tools to create sexualized versions of photos showing real children. The apps make it easy for someone to take an ordinary picture and change it without the person's knowledge or consent.
According to the Tech Transparency Project, TTP matched "multiple CSAM ads to photos of real children that appeared online." This means the images in the ads were not just computer-generated faces — they were actual children whose pictures had been taken from public sources and manipulated.
Real Children Targeted in the Ads
The investigation identified specific cases where real children were victimized. One ad used a press photo of a young member of a European royal family. Another used images swiped from a popular Instagram profile of a preteen girl who was considered an influencer.
These ads appear to violate federal child pornography laws. When real children are depicted in manipulated sexual content, it crosses a legal line that goes beyond typical content moderation failures.
"Multiple CSAM ads to photos of real children that appeared online." — Tech Transparency Project
Meta's Slow Response to Reported Ads
The report highlights a serious gap in Meta's content moderation systems. Despite having automated tools and human reviewers, the company took days to remove ads that clearly contained child sexual abuse material. This delay meant the ads remained visible to users for extended periods.
The fact that these ads were promoting apps — not just spreading harmful content — adds another layer of concern. Meta was essentially running advertising revenue from companies whose products enable child exploitation.
Why This Matters for Platform Safety
This investigation raises important questions about how social media platforms handle AI-generated abuse content. As AI tools become more powerful and accessible, the risk to real children grows. Platforms need faster detection systems and clearer policies for AI-manipulated imagery.
The report also points to a broader issue: advertising systems that prioritize revenue over safety. When ads slip through for days, it suggests the review process is not working as it should.
Our Take: Meta Must Act Faster on AI Abuse Content
To put it plainly, this report shows a disturbing failure on Meta's part. Taking days to remove ads that sexualize real children is not acceptable. These are not borderline cases — they are clear violations of both platform policy and federal law.
The use of real children's photos makes this especially serious. A young royal family member and a preteen influencer did not ask for their images to be used this way. Their photos were taken from public sources and weaponized without consent.
In our view, Meta needs to treat AI-generated CSAM with the same urgency as traditional CSAM. The company should invest in better detection tools specifically designed to catch manipulated imagery. It should also review how ads from AI app developers are vetted before they go live.
Parents and guardians should also be aware that public photos of children — even on seemingly harmless profiles — can be scraped and misused. This is not about blaming victims; it is about understanding the risks in an age where AI can alter any image.
The Tech Transparency Project's findings should push Meta and other platforms to act. Every day an ad like this stays up is another day a real child is exploited.
