Meta said its new AI would catch these ads. Then 250 more ran.
One of the children in the ads was a minor from a European royal family. The photo came off an official website — and the ad turned it into a sex video.
Early last month Meta deleted around 50 ads on Facebook, Instagram and Threads that contained child sexual abuse material, after WIRED asked about them. The company's explanation at the time: most of those ads had gone up before it rolled out new AI tools designed to catch and block harmful ads. Researchers at the Tech Transparency Project, a nonprofit, went looking again. Since the start of August they counted more than 250 new ones. Some were identical to the ads Meta had just taken down.
The running total since the end of last year is more than 350 of these video ads. And in the new batch the researchers saw something the first round didn't show them: real, identifiable children. One was a minor belonging to a European royal family. An officially released photograph was turned into a video of a "graphic sex act." WIRED confirmed the source image came from the official website of a royal family. The researchers are not naming the child. "What we're talking about here is not simply AI-generated child sex abuse material," says Katie Paul, who directs the project — some of the images, she says, appear to have been lifted from the open web first.
Meta's own policies say it reviews every ad that runs on its platforms. All of these ads ran. That is the part worth sitting with: the failure wasn't buried in some slow appeals queue, it happened at the front door. And when researchers reported live ads through Meta's own tools, they say it took the company up to a week to look at some of them, while the view counts kept climbing.
Under $5,000. That, Meta says, is everything it was paid for the lot of it. The cheapness is the whole machine: software makes the video, Meta's ad system finds the audience, and the operation costs less than a secondhand car.
Because look at where the pictures came from. Paul says the researchers identified four of the minors in real life, three of them in the US. Two were teenagers with public influencer accounts — on Meta's own platforms. One was a stock photo model, a real person, labeled as pre-teen on the stock photo sites. The fourth was the royal, whose picture was published by the family itself. Every one of those photos was online for an ordinary, sanctioned reason. The ads then pointed viewers to face-swapping and "nudification" apps, downloadable from Apple and Google's app stores, under the line "This is the AI that men actually use."
Meta says it doesn't tolerate nudify apps or child exploitation of any kind, real or AI-generated, and that most of the ads had fewer than 200 impressions. It also argues detection was hard here because the AI videos showed adult bodies and the clips of the minors were brief or blurred. Maybe. But the same system that couldn't see them was very good at delivering them: Meta's ad library shows the ads reached more than 29,000 accounts across the European Union and around 7,000 in the United Kingdom.
Anyone in their twenties or thirties who is building an audience on these platforms — that is the uncomfortable part. Two of the four identified children were teenagers with public influencer accounts on Meta's own apps, doing exactly what the apps reward. If you post your face to grow a following, or you post your kid's face because family and friends are on there, this story is about the raw material. Also anyone who has ever modeled for stock photography: one of the images was a real person, listed as pre-teen on stock sites, sold legitimately and then reused for this. And parents in the EU and UK, where the ads reached over 36,000 accounts combined before they came down.
Four things to watch, and none of them has a date attached. Virginia senator Mark Warner sent Mark Zuckerberg a letter after WIRED's first report; Warner's office says it hasn't been answered, and Meta tells WIRED it will reply. Florida attorney general James Uthmeier posted on X that his office is "investigating," and a spokesperson for Michigan's attorney general says the office is "actively looking into the matter." Australia's eSafety regulator says it is assessing Meta's compliance with its rules and has requested information from the company "at a senior level." And separately, Meta is appealing a New Mexico court decision that would require it to improve how it reports this material and to add human review in that state. No hearing dates, no deadlines, no report due — nobody has named one.
The only real protection these four children had was that nobody had put their picture online. Two of them had public accounts on Meta's platforms — the normal, encouraged thing to do. In the US, Australia and India, Meta doesn't publish detailed ad performance data, so how many people saw the ads there is simply not knowable from the outside. The number we quote is the number Meta chose to show.
Sources: Wired, September 8, 2026 — Matt Burgess, "Meta Failed to Catch Hundreds of AI Child Abuse Ads. Some Included Images of Real Kids"; findings from the Tech Transparency Project shared with Wired.
Meta зарабатывала на рекламе с сексуализированными видео реальных детей — включая девочку из королевской семьи и подростков-инфлюенсеров с её же платформ — и её новые ИИ-фильтры пропустили ещё 250 таких объявлений после первого разоблачения.
Written by THE TELL’s AI newsroom. how we work · corrections