Reports of child sexual abuse material linked to generative AI have exploded from roughly 4,700 in 2023 to more than 400,000 in 2025 — an increase of over 8,400% in two years, according to data from Childlight, the Global Child Safety Institute based at the University of Edinburgh.
The figures were published on Monday, November 16 as part of Childlight's 2026 Into the Light Index and reported by The Guardian. They represent referrals that technology companies filed with the US National Center for Missing and Exploited Children (NCMEC) — a count of what platforms detected and reported, not a census of everything in circulation. The surge is the clearest quantitative signal yet that generative AI abuse, long described by the industry as a risk on the horizon, has become one of the largest categories of online child exploitation reporting, a shift documented in ongoing breaking AI news coverage of platform safety and regulation.
From 4,700 to 400,000 Referrals
The index tracks reports that technology companies are legally required to send to NCMEC when they detect suspected child sexual abuse material on their services. The jump from about 4,700 referrals in 2023 to more than 400,000 in 2025 maps almost exactly onto the consumer rollout of image generators and cheap fine-tuning tools over the same period.
The headline figure is not the only number in circulation. NCMEC has said separately that its CyberTipline received more than one million generative-AI-related reports between January and September of 2025 — a larger total than the 400,000 Childlight cites, reflecting differences in how the two counts are scoped. The two datasets measure overlapping but distinct things, and both point the same direction.
Nudify Apps and Deepfakes of Real Children
Childlight's report also highlights the growth of so-called nudify apps, which use generative AI to place a child's face onto explicit imagery. The pictures are synthetic, but the harm is not: the images depict real children, are shared among offenders, and are increasingly used for blackmail and harassment of minors.
Professor Debi Fry, Childlight's director of data, said the speed at which AI is changing the threat to children is deeply concerning, according to the Guardian's reporting. The concern is no longer hypothetical capability but industrial-scale production: what required effort and skill in 2023 now requires only an app store download.
What the Watchdogs Are Finding
Independent assessments corroborate the scale of the problem. The Internet Watch Foundation, the UK watchdog that directly assesses reported material, found 6,310 AI-generated images that met the legal definition of child sexual abuse in the first half of 2026 — 40% more than in all of 2025.
The trajectory for video is starker still. The IWF identified 3,440 AI-generated child sexual abuse videos in 2025, up from just 13 in 2024, and 65% of them were in Category A, the most severe grade under UK law. A jump from 13 to 3,440 in a single year shows the tools moved from still images to moving video faster than detection systems could adapt.
Regulation Is Racing to Catch Up
The UK has moved furthest so far. Under the Crime and Policing Act 2026, it is now a criminal offence to possess, create or distribute AI tools designed to generate child abuse material, punishable by up to five years in prison, according to the government's own factsheet. The same legislation criminalises making or supplying nudification tools — a provision aimed directly at the app developers behind them.
Elsewhere the picture is thinner. Childlight's index found that only 22 of the 103 countries it examined have comprehensive survey data covering both offline and technology-facilitated child abuse, meaning most governments are legislating without a reliable baseline. The institute is calling for tougher online safety laws, better data collection, greater accountability from the technology sector and earlier prevention work.
The Reporting Pipeline Problem
One limitation matters when interpreting the numbers: the data does not name specific models or platforms as the source of the reported material. The tips come from technology firms that scan their services and report what they find — which means a company that does not scan does not generate reports. Detection-led reporting is, as the Guardian's reporting notes, a thin kind of safety, and the next set of figures will show which companies are actually looking.
For the AI industry, the message of the index is unambiguous. Abuse of generative tools is no longer a projection in a model card's risk section; it is the biggest line in the ledger, and the companies whose models and apps made the abuse cheap are about to be asked to pay for it — in enforcement actions, in compliance costs, and in law.
Stay Ahead of AI
Follow the debate over AI safety and regulation as it unfolds. Read more AI news →
