Researchers analyzing dark web forums found evidence of growing communities dedicated to using AI to generate child sexual abuse imagery. The Internet Watch Foundation documented 20,254 AI-generated images uploaded to one dark web forum in October 2023 alone, and the numbers have continued to rise. The UK government plans new laws making it illegal to possess, create, or distribute AI tools designed to generate CSAM, punishable by up to five years in prison. Paedophile manuals teaching the use of AI for child abuse will also be criminalized.
What the Dark Web Research Found
Bailey and Lundrigan’s team analyzed forum chats from dark web spaces over 12 months. What emerged was not simply a record of images being shared; it was evidence of an active, educating community.
Forum members used existing non-AI images to train AI models, hoping to make better images over time. They shared tips on which tools worked best, talked about avoiding detection, and encouraged others to learn these skills. They even called people who made these images “artists,” treating child exploitation like a creative activity.
The researchers also documented a clear belief among many forum users that AI-generated imagery is a victimless crime because the children in the images aren’t real. That belief is the central myth the research challenges.
Why “AI-Generated” Does Not Mean “Harmless”
The belief that fake abuse images do no harm is wrong for several reasons.
First, the images are often not entirely fictional. Many AI image generators require training data. In the context being documented on dark web forums, offenders were using existing abuse imagery, including photographs of real children, as the foundation material to teach models how to generate new content.
The real children in those original images remain victims each time derivative content is created. Second, the research from wider child protection literature is clear about how images, real or manufactured, are used in the cycle of abuse. Offenders use visual material to groom children, normalizing what they’re shown. They use it as blackmail. The existence of imagery, even where a specific child wasn’t directly photographed for a given image, extends harm into new territories rather than eliminating it.
Third, there is a human cost to those who review it. Moderators and investigators required to examine abuse imagery in detail to determine whether it is AI-generated or photographed face serious occupational trauma. This reality is documented within law enforcement and the charitable sector, and the growing volume of AI material makes it more acute.
The Internet Watch Foundation’s Numbers
The Internet Watch Foundation (IWF), the UK charity that identifies and removes online abuse imagery, has published data that tracks the growth trajectory.
In October 2023, the IWF documented 20,254 AI-generated images uploaded to a single dark web forum in one month alone. Before that report, this scale of AI-generated abuse imagery on dark web forums had not been publicly documented. Since then, the IWF reports that while AI-generated material remains smaller in total volume than non-AI imagery, it is growing significantly faster.
Placed alongside the IWF’s finding of an 830% rise in online child sexual abuse imagery overall since 2014, the AI trajectory is one of the most concerning developments in child protection in recent years.
Understanding where this material circulates requires understanding the dark web itself: encrypted, hidden, accessible only through software like Tor, and designed to resist the kind of identification and surveillance that apply on the open internet. We’ve covered how law enforcement approaches this infrastructure in depth. Still, the sheer volume of material across thousands of forums makes comprehensive monitoring an ongoing challenge, even for well-resourced agencies.
What the UK Laws Would Do
The UK government’s proposed legislation has three main components.
It would make it illegal to possess AI tools specifically designed to generate CSAM, not just to possess the images themselves, but to own the means of production. This is the key innovation. Current UK law criminalizes indecent images and pseudo-photographs of children. What it does not cover is someone in possession of a configured AI tool who has not yet generated an image. The new law closes that gap, meaning investigators could act against someone even if no images are found on a device at the time of a search.
It would criminalize “paedophile manuals”, instructional material teaching others how to use AI to create CSAM or facilitate child abuse more broadly. These are educational resources for perpetrators, and the Anglia Ruskin research documented their existence in dark web forums.
The maximum sentence for offenses related to possessing or distributing AI tools designed for this purpose would be five years in prison.
The legislation also sits alongside existing government plans to criminalize the creation of deepfake pornography involving adults, reflecting a broader legislative response to AI-generated harmful content.
The Case for Speed
Bailey and Lundrigan’s analysis ends with a warning worth taking seriously: law enforcement will always lag behind offenders in adopting new technology. The window for legislation to meaningfully restrain the spread of AI-generated CSAM tools is narrowing as the tools become cheaper, easier to use, and more widely distributed.
Each month legislation is delayed, more AI toolkits circulate on dark web forums, more training pipelines improve, and more individuals who might otherwise hesitate are shown by online communities how to begin. The researchers frame this as a systemic risk. Without laws that reach the tools themselves, not just their outputs, investigators will always respond to images after they exist rather than intercept capability before it’s created.
The UK becoming the first country to legislate specifically against AI-generated CSAM tools sets a precedent other countries will likely follow. But on its own, it doesn’t address the global nature of these forums, which host members across dozens of countries. Effective enforcement will require international coordination of the kind that has driven the most successful dark web marketplace and CSAM platform takedowns; domestic legislation is a necessary condition, not a sufficient one.
The Broader Technology Design Argument
Bailey and Lundrigan also say the problem is bigger than law enforcement. Many AI image tools were made for good reasons, but offenders can use them for harm.
This leads to a design question for AI developers and policymakers: can tools be built to prevent misuse from the start, instead of fixing problems later? The researchers suggest a whole-system approach that begins when new technology is designed. They want the tech sector to take responsibility for possible misuse before products are released, not after harm happens.
The connection to the broader AI safety debate is clear: AI capability without adequate guardrails creates risks that fall on the most vulnerable people, not just on enterprises or governments.
How to Report
If you find child sexual abuse material online or are worried about a child’s safety, there are ways to report it.
In the UK, report to the Internet Watch Foundation at iwf.org.uk/report or to the CEOP Command (Child Exploitation and Online Protection) at ceop.police.uk.
In the United States, file a report with the National Center for Missing & Exploited Children via the CyberTipline at missingkids.org.
Frequently Asked Questions
Is AI-generated child abuse imagery illegal in the UK?
Yes, existing UK law already criminalizes AI-generated pseudo-photographs of children. The new proposed legislation goes further, making it illegal to possess or distribute the AI tools designed to create CSAM, even if no images have been generated yet.
What did the Anglia Ruskin research find on dark web forums?
Researchers found active communities using AI tools to generate child sexual abuse imagery, sharing tips, training models on existing material, and describing image creators as “artists.” Members hoped the technology would become easier to use over time.
How many AI-generated images did the IWF find?
In October 2023, the Internet Watch Foundation documented 20,254 AI-generated images uploaded to a single dark web forum in one month. Numbers have continued to grow since then, though AI material still represents a smaller share of total CSAM volume than non-AI images.
Is AI-generated CSAM really harmful if no real child was photographed?
Yes. The images are often trained on real abuse material, meaning real children are still part of the pipeline. Additionally, the imagery is used to groom, blackmail, and normalize abuse , the harm doesn’t require a direct photographic victim to be real.
What prison sentences would the new UK laws carry?
Up to five years for possessing, creating, or distributing AI tools designed to generate CSAM. The legislation would also criminalise paedophile manuals teaching AI-based abuse methods.
Where do I report CSAM online?
In the UK: the Internet Watch Foundation at iwf.org.uk/report or CEOP at ceop.police.uk. In the US: the NCMEC CyberTipline at missingkids.org.
