AI tools are now creating realistic child sexual abuse material much faster than before. The IWF found more of this content in the first half of 2026 than in all of 2025. Police have trouble telling AI-generated images from real ones, which slows down help for real victims. Right now, the main defenses are new detection technology and working together across countries.
How AI Creates This Material
Understanding the threat means understanding the methods (not in any way to replicate them) and in doing so you understand why the existing detection tools are so poor at detecting it.
The US Department of Homeland Security has documented the main ways AI is misused to create this material. These include generating entirely fictional content depicting children, using deepfake technology to alter innocent images, and de-aging tools that manipulate adult images. Guides on how to use these AI tools have appeared on the dark web, as we covered in our piece on Operation Alice. This international operation dismantled over 373,000 dark web sites used for distributing child sexual abuse material.
Detective Superintendent Frank Rayner (from Australia’s CyberSecurity Cooperative Research Centre) put the issue of accessibility down very plainly: “The tools that people can access online to create and edit using AI are proliferating and getting more sophisticated. You can just go onto a browser, enter your prompts and have a result in minutes.”
That accessibility feeds demand. According to Thorn, a child safety technology company, AI doesn’t just create more material; it creates material faster than existing detection systems can classify it, generating new content that hasn’t been cataloged in known-image databases.
Why Detection Has Become So Much Harder
Traditional detection of child sexual abuse material depends on recognizing known images. When a file matches a previously identified hash, a digital fingerprint, it can be flagged automatically. This works for material that already exists in databases.
AI-generated content bypasses this system entirely. Every AI-generated image is new and unique. It has no hash in any existing database. Automated systems built to catch known content see nothing. Human analysts must review it individually.
This creates two simultaneous problems for law enforcement. First, it dramatically increases workload. Second, it makes it harder to distinguish AI-generated material from real footage of actual abuse occurring to a real child.
Australian Federal Police Commander Helen Schneider described the operational consequence: “It’s sometimes difficult to discern fact from fiction, and therefore we can potentially waste resources looking at images that don’t actually contain real child victims. It means there are victims out there that remain in harmful situations for longer.”
That final sentence is the one that counts. The analysts are busy with the synthetic content, the real children are still out there waiting to be identified. The amount of synthetic content doesn’t in any way offset the real abuse, it just sits on top of it and smotheres it.
The same issue arises with identification. Thorn’s work has highlighted that AI can actually make identification of the victim more difficult. If an offender uses an AI tool to create new material featuring a real child without in any way capturing the actual abuse then the identity of that child may be more difficult to trace through the AI generated version than it would be through the original recording.
This is why AI used defensively, trained on dark web content to detect threats including abuse imagery, has become a genuine research priority. The same multimodal AI systems that can read dark web forums for threat intelligence are now being trained to identify AI-generated abuse content by looking for subtle technical inconsistencies that human reviewers can’t reliably spot.
The Scale in Numbers
To understand why this is being treated as a crisis the numbers from a number of different jurisdictions tell the same story.
In the US, the National Centre for Missing & Exploited Children (NCMEC) received over 440,000 reports of AI-generated material relating to child sexual exploitation in the first half of 2025 alone. That’s in comparison to the roughly 7,000 reports received across the entirety of 2023 and 2024. That’s a massive shift in the threat landscape in just two years.
The IWF’s 2026 report estimated that at least 1.2 million children had their image used to generate this type of content.
In Australia (which the CyberSecurity Cooperative Research Centre has identified as the 3rd largest market in the world for online sexual abuse material) the Australian Centre to Counter Child Exploitation received over 49,500 reports in the 2023-2024 financial year, some 9,300 more than the year before.
AI generated content still represents around 1% of all child sexual abuse material that we detect. Its growth rate is such that that percentage is increasing very quickly and the legislative and technical frameworks to deal with it have not kept up.
What Responses Are Being Built
The IWF has pushed the EU to legislate so that platforms are required to detect both known and previously unseen abusive content (not just content that’s already in their databases). Their CEO went so far as to state that “a legal framework that is solely focused on known material would leave the door open for offenders to generate and disseminate vast quantities of new AI material very quickly”.
The Council of Europe has taken a tangible step in this area by criminalizing the creation, modification and distribution of AI-generated child sexual abuse material under its Conventions. In the UK, the creation or distribution of AI models designed to generate such material is now a criminal offence and carries a potential sentence of five years.
Europol led Operation Torch in July 2026 with 28 arrests across seven EU countries as part of a concerted effort to target the distribution networks of AI generated child sexual abuse material. These operations take months to build and are very resource-intensive and simply can’t keep up with the rate at which AI tools can generate new content.
In 2024 Google, Meta and Amazon formed a specific alliance to build technical defences against the misuse of AI for this purpose. Thorn has always maintained that any real solution requires AI developers, data hosting platforms, social media companies and search engines to be on board from the outset rather than just reacting after the material has already been out there.
How to Report It
Reporting is one of the most important tools in the public’s armory.
The IWF in the UK and abroad takes reports of child sexual abuse material (CSAM) seriously. When they confirm material, they provide the hash to the relevant law enforcement and tech companies.
In the United States, the NCMEC CyberTipline accepts reports from anyone who encounters this content online, including material that appears AI-generated.
In the UK, CEOP (Child Exploitation and Online Protection) accepts reports and guides children, parents, and professionals.
In Australia, reports go to the Australian Center to Counter Child Exploitation (ACCCE) at police.gov.au.
Both the UK National Crime Agency and the IWF have also advised parents and guardians to keep children’s photos off public social media where possible; AI tools can use publicly available images as source material. Making accounts private significantly reduces this risk.
Frequently Asked Questions
What is AI-generated CSAM?
Child sexual abuse material created via AI (deepfakes, de-aged images and fully synthetic content etc). It’s illegal in most jurisdictions regardless of whether or not a real child is actually filmed and it causes real and ongoing harm.
How much has AI-generated CSAM grown?
The IWF assessed 6,310 AI generated images (meeting the legal definition of CSAM) in H1 2026 alone which is 40% more than the 4,512 they assessed across all of 2025. NCMEC received over 440,000 reports in the first half of 2025 alone as opposed to the roughly 7,000 they received across all of 2023-2024.
Why is it harder for police to detect?
AI-generated images are new and unique every time. They have no digital fingerprint in existing databases. Law enforcement must review them manually, which consumes time and resources that would otherwise go toward identifying real victims of real abuse.
Is AI-generated CSAM illegal?
Yes. In the UK you can be sentenced to up to 5 years in prison for creating or distributing AI models of this type. The Council of Europe has criminalised the creation, modification and distribution of such models. The US, Australia and the majority of EU countries have similar legislation in place or in the process of being developed.
