news

IWF assessed 6,310 AI-generated child abuse images in six months, 40% above all of 2025

The figure covers still images that analysts found and classified between January and June, not videos, and not an estimate of how much material is circulating.

By Daniel Ochoa · 7 October 2026

The Internet Watch Foundation assessed 6,310 AI-generated images that met the legal definition of child sexual abuse between 1 January and 30 June 2026, 40% more than the 4,512 its analysts identified across the whole of 2025. The six-month count has already passed the previous full year, and the charity published it alongside a call for EU policymakers to agree long-delayed legislation that would give technology companies lasting legal certainty to find and remove criminal material on their platforms.

Girls appear in 98% of the images where age and gender were both recorded

Of the 6,310 images, analysts recorded both age and gender for 6,221. Girls featured in 98% of those. The number of unique images depicting girls rose from 4,259 across all of 2025 to 6,094 by the end of June this year. The age breakdown for the first half of 2026 covers 2,534 images depicting children aged seven to 10, 2,369 depicting children aged 11 to 13, 1,004 depicting children aged three to six, and 190 depicting infants and toddlers under the age of two. Children aged seven to 13 accounted for 79% of the AI-generated imagery identified in the period, up from 70% across 2025.

The number counts assessments, not what exists online

This is a detection figure, not a prevalence figure. Every one of the 6,310 images is something an analyst located, viewed and judged against a legal definition, which means the total is bounded by how much work a hotline can physically do in six months. A 40% rise can reflect more material, more effective discovery, or both, and the release does not separate those. The scope is also narrower than the headline suggests, because the count covers still images and excludes AI-generated video entirely.

Hash matching cannot catch an image nobody has seen yet

The IWF converts confirmed illegal images into hashes, digital fingerprints that technology companies and law enforcement can use to recognise and block known material. That method works on files that have already been assessed at least once. Generative models produce novel output on demand, so every new image starts outside the hash list and stays undetected until an analyst finds it and adds it. The detection stack is built around matching, and the supply side is built around never repeating itself.

The policy ask points upstream at the model builders

The IWF frames the figures as an argument about how models are built rather than only about how platforms moderate. Its position is that online services currently lack a durable legal basis to detect and remove this content, including material that has not been seen before, and that the gap is a legislative one. The same argument applies to the generation side: if the only reliable control is classification after the fact, the burden sits permanently with human reviewers working through a backlog that a model can refill in seconds.

For anyone tracking how AI-generated imagery gets scored and classified, the useful detail here is methodological. The IWF publishes what it assessed, over what window, with age and gender breakdowns and an explicit note on what the figures exclude. That is a disclosed method with stated limits, which is more than most published numbers about generated media come with, and it is why the 6,310 figure is worth reading as a floor rather than a measurement of the problem.

Sources: Internet Watch Foundation, National Technology.

Daniel Ochoa: Covers model launches, shutdowns and pricing changes as they happen. Reads deprecation notices for a living so you do not have to.