Scroll through Facebook and you expect ads for shoes, teeth whitening, and perhaps a conspiracy theory. You don’t expect to see child sexual abuse imagery. But researchers from The Transparency Project found more than 50 ads containing AI-generated CSAM that ran on Meta’s platforms. Fifty. Ads. Paid to reach your eyes.
Meta’s response? \”We removed them.\” Thanks. But they’d already been seen. And the tools that were supposed to catch them—the same \”automated tools\” that approve billions of ad images every year—had waved them through. This isn’t an anomaly. It’s the inevitable product of a company that outsourced judgment to algorithms.
This is not a moderation failure. This is the algorithm acting on its true instruction: maximize engagement, no matter the cost.
Think about how an ad gets onto your feed. An advertiser uploads an image. Meta’s system scans it for rules. The scan happens in milliseconds, because there are millions of ads. Speed is king. Volume is king. Context is a nuisance. So the system asks one question: will this drive clicks? It doesn’t ask: does this violate the law?
That’s the twist nobody wants to talk about. We like to believe that Meta is trying to keep us safe but failing. The evidence points elsewhere: the ad auction is a meritocracy of malevolence. Every image, legitimate or criminal, is treated as inventory. The best-performing ad wins. And AI-generated CSAM can perform incredibly well, because it’s designed to trigger attention.
The company calls this an \”ongoing battle.\” But it’s not a battle. It’s a business choice. Automated moderation is cheaper than human review. That cost-cutting is passed on to us as policy. And when an image slips through, the response is the same: a quiet removal, a statement of \”commitment to safety,\” and no actual change.
The only thing more profitable than your attention is the crisis that distracts from how it’s being sold.
This is what happens when an algorithm’s success metric is engagement, and the only stakeholder is the advertiser’s wallet. The child in the image is not a person. The child is a placeholder. The ad system doesn’t recognize abuse because it was never built to recognize abuse. It was built to recognize thumbs-ups.
If you use Instagram, Facebook, or WhatsApp, this story is about you. You’re not a user. You’re an audience member in a live auction where the highest bidder gets your eyeballs. Sometimes the bidder is selling a mattress. Sometimes the bidder is selling the sexualized image of a child. The line between those two is not drawn by a human. It’s drawn by a cost-benefit analysis.
And here’s the uncomfortable truth: as AI image generation improves, this will get worse before it gets better. The same tools that create lifelike abuse are being refined daily. Meta’s automated systems are not merely failing—they’re being outpaced. The only solution is to slow down, hire humans, and accept that safety costs money. But slowing down is the one thing Meta’s business model cannot do.
Meta doesn’t have a safety problem. It has a profit problem. And your outrage is the only moderation system it still fears.
So don’t let them tell you it’s fixed. It isn’t. The ads are gone, but the algorithm that let them through is still running. And it’s learning.
FAQ
Q: What is the key takeaway?
A: See the article.