In a one-month snapshot of a single dark-web forum, analysts at the Internet Watch Foundation recorded 20,254 AI-generated child sexual abuse images. The files were synthetic, but the damage they cause is not hypothetical.

What that looks like on the ground appeared in Almendralejo, Spain, in September 2023. A mother there — call her Mariana — opened a picture circulating in a school group chat and saw her own daughter's face placed on a naked body. No camera had taken that photo. It had been generated with an app from a clothed portrait. Within days, local police were investigating fake nudes of more than 20 girls between 11 and 17.

That cluster explains why researchers now treat AI-generated sexual images as a new delivery system for an old form of abuse, not a future projection.

Synthetic pixels, real damage

Nonconsensual intimate imagery, or NCII, describes intimate pictures or video shared without permission. AI-generated sexual images belong under this umbrella, even when the person shown never undressed and the nude was never photographed. The only raw material needed is a face.

Generative models have collapsed the barrier. Undress apps take a school portrait or a social media photo and return a convincing nude in under a minute. Text-to-image tools can produce child sexual abuse material from a prompt alone, with no recorded victim. What makes these images intimate is not how they were made; it is that they reproduce a recognisable person without consent.

The numbers behind the surge

The Internet Watch Foundation's count of 20,254 AI-generated child sexual abuse images in one month did not come from a laboratory prediction. Analysts traced the images to a single dark-web forum and classified thousands as criminal. The same organisation says the share of reports involving AI content has moved from occasional to routine.

An earlier benchmark from Deeptrace set the pattern for the current problem: 96 percent of deepfake videos online in 2019 were non-consensual pornography, and women appeared in 99 percent of them. The tools have improved since then, but the target profile has not shifted.

The more recent shift is one of skill. Free and open-source models run on ordinary laptops, and forums trade model weights and prompts the way earlier communities traded stolen photos. Helplines now describe batches of images produced from two or three source photos instead of a single leaked file. The Internet Watch Foundation publishes regular public data on reporting and removals.

Photo by Ahtziri Lagarde on Unsplash

Why fake images injure real people

The phrase 'it's not real' misunderstands image-based sexual abuse. A victim does not experience the file as fake. She sees her face on a body she never posed for, and she knows the image existed in another pupil's group chat before she knew it existed.

In Almendralejo, some girls stopped attending class. Parents described daughters who would not leave the house, refused to appear in photographs, or asked whether they should change schools. For adults, the same pressure lands differently. Female journalists, streamers, human rights researchers and activists describe deleting public accounts or reducing bylines because a fake sexual image can surface in a hiring check, a security clearance, or a family argument. The image may be made of mathematical noise, but its consequences are concrete: lost income, lost trust, nights of broken sleep, and a smaller public life.

Data protection authorities and abuse services report that victims often show panic attacks, hypervigilance about their own photo, self-surveillance and fear of ordinary phone cameras. Some stop appearing in video calls or remove their photograph from work pages. The clinical picture overlaps with the harm documented for photographed abuse, because the social meaning of the image is not virtual.

The legal response is a patchwork

No single global law covers all cases. In the United Kingdom, the Online Safety Act made sharing an intimate image without consent an offence, including images that have been digitally created. The European Union's Digital Services Act requires very large online platforms to act on reports of nonconsensual intimate imagery. Australia's eSafety Commissioner can order removal of intimate images and has extended guidance to synthetic sexually explicit material. Across the United States, dozens of state laws address the problem differently, and a federal proposal called the TAKE IT DOWN Act has drawn bipartisan backing without closing every gap.

Platform policy moves faster than legislation in some places. Large adult sites and search engines have created reporting routes for nonconsensual content, and some hubs for deepfake images were removed after payment processors refused to work with them. Abuse groups caution that moderation remains inconsistent: an image removed in one hour on one platform may remain indexed elsewhere for months.

What the evidence still cannot settle

Researchers are still working on detection systems that can separate AI-generated images from real photographs. Current tools miss some images and flag others, and a detector is only useful if platforms choose to run it. Watermarking proposals face similar limits: a visible watermark can be cropped, and invisible forensic markers degrade when the file is recompressed.

The bigger open question is measurement. Dark-web counts understate the problem because they capture only forums researchers can see. Survivor surveys show the same issue from the other side: people who report may be a small fraction of those affected, and many never tell anyone because the image itself is humiliating.

Photo by Markus Winkler on Unsplash

What to do in the first 24 hours

Victim support services return to the same sequence because it preserves a person's options. If the person in the image is you or your child, resist the urge to delete every trace before documenting it. A deletion can destroy the evidence a platform or police officer needs to act.

• Save URLs, timestamps, account handles and message IDs in a folder outside the platform. Include screenshots of search results if the image has spread.
• Report each post through the platform's reporting tool and keep the reference numbers. Flag the content as nonconsensual intimate imagery, not as spam.
• Get a digital fingerprint of the image. Adults can use StopNCII.org, which hashes images so participating platforms can block them without the file leaving your device.
• Contact police and a local support line, especially if someone is demanding money or more images. Do not pay a blackmailer.

For people under 18, the National Center for Missing and Exploited Children runs a parallel removal service called Take It Down. Parents should avoid shaming the child; the important point is that the image was made without consent, no matter what the child originally shared.

If you do only one thing in the next hour, open StopNCII.org if you are an adult, or Take It Down if the person in the image is under 18, and start the hash process.