The numbers are staggering, however specialists say what we’re seeing is barely the start. As AI-generated youngster sexual abuse materials, or CSAM, surges to document ranges, researchers warn that the expertise isn’t simply producing extra dangerous content material, however it’s essentially altering how kids are focused; how survivors are revictimized; and the way investigators are overwhelmed.
Investigators already had their arms full with scrubbing CSAM from the web. However with generative AI, that problem has been exacerbated. The Web Watch Basis (IWF), Europe’s largest hotline for combating on-line youngster sexual abuse imagery, documented a 260-fold improve in AI-generated youngster sexual abuse movies in 2025. It went from simply 13 movies the yr prior to three,443. Researchers who’ve spent years monitoring this problem say the explosion just isn’t a shock. It’s, nonetheless, a warning.
“Any numbers that we see, it’s the tip of the iceberg,” stated Melissa Stroebel, vp of analysis and strategic insights at Thorn, a nonprofit that builds expertise to fight on-line youngster sexual exploitation. “That is about what has been either detected or proactively reported.”
The surge is a direct consequence of generative AI turning into quicker, cheaper, and extra accessible to dangerous actors. Thorn has recognized three distinct methods these instruments at the moment are being weaponized towards kids.
The primary is the revictimization of historic abuse survivors. A toddler who was abused in 2010 and whose photos have circulated on-line for over a decade now faces a completely new layer of hurt. Offenders are utilizing AI to take these present photos and personalize them: inserting themselves into recorded scenes of abuse to provide new materials.
“In the same way that you can Photoshop Grandma who missed the Christmas picture into the Christmas picture,” Stroebel advised Fortune, “bad actors can Photoshop themselves into scenes and records of an identified child.” That course of creates contemporary victimization for survivors who might have spent years making an attempt to maneuver previous their abuse.
The second is the weaponization of harmless photos. A photograph of a kid on a faculty soccer staff internet web page is now potential supply materials for abuse. With broadly obtainable AI instruments, an offender can convert that totally benign picture into sexual abuse materials in minutes. Thorn can also be documenting peer-on-peer instances, the place an adolescent generates abusive imagery of a classmate with out totally greedy the severity of the hurt they’re inflicting.
The third, and most systemic, affect is the pressure being positioned on already overwhelmed reporting pipelines. The Nationwide Heart for Lacking and Exploited Youngsters receives tens of tens of millions of CSAM reviews yearly. The velocity with which AI can now generate novel materials dramatically compounds that burden and creates a brand new urgency. When a brand new picture arrives, investigators should decide whether or not it depicts a toddler in lively hazard proper now, or is an AI-generated picture.
“Those are really critical inputs to help them triage and respond to these cases,” Stroebel stated. AI-generated content material makes these determinations considerably tougher, however she added each instances of a picture taken in actual time and an AI-generated picture are reported and handled the identical approach by authorities.
The expertise has additionally made a number of the most repeated youngster security steerage dangerously outdated. For years, kids have been warned to not share photos on-line as a fundamental safeguard towards exploitation. That recommendation not holds. Thorn’s personal analysis discovered that one in 17 younger folks have personally skilled deepfake imagery abuse, and one in eight knew somebody who had been focused. Victims of sextortion at the moment are being despatched photos that look precisely like them—photos they by no means took.
“There’s no need for a child to have shared an image any longer for them to be targeted for exploitation,” Stroebel stated.
On the detection entrance, conventional hashing expertise, which works like a digital fingerprint for identified abuse information, can’t establish AI-generated content material as a result of every synthetically created picture is technically new. Take, for instance, a photograph of one thing very well-known, just like the Statue of Liberty. That photograph of the statue has a digital fingerprint. Now, say you zoom in, zoom in some extra, and zoom in once more to alter the shading of 1 pixel by 0.1%. That change is probably going imperceptible to the human eye. Nonetheless, the fingerprint of that photograph is now utterly new, which means the hashing expertise doesn’t acknowledge it as the identical photograph with simply that one pixel distinction.
Beforehand, underneath conventional hashing expertise, making that one pixel distinction to a photograph identified to be CSAM would imply it will go undetected by the tech. Nonetheless, classifier expertise, which evaluates what a picture incorporates relatively than matching it to a identified file, is now important to catching content material that might in any other case slip by way of totally.
For fogeys, Stroebel’s message is pressing and unambiguous. The dialog can’t wait, and it should go additional than previous warnings. If a toddler comes ahead, the primary response can’t be skepticism: “Our job is, ‘Are you safe, and how do I help you move through to the next step?’”
