IWF counts 6,310 illegal AI images in just six months
October 5, 2026
The Internet Watch Foundation found 40 percent more AI-generated child sexual abuse images in the first half of 2026 than in all of 2025. New images can evade known hash blocks.
What this is about
The Internet Watch Foundation (IWF) published new figures on AI-generated child sexual abuse material on October 5, 2026. Its specialists assessed 6,310 AI-generated still images as illegal between January 1 and June 30. That is 40 percent more than the 4,512 images recorded during the whole of 2025.
The figure is not an estimate for the entire internet. It covers only images submitted to the UK-based hotline and legally assessed by trained specialists. Even so, the comparison shows how quickly new variants can be produced. The IWF is therefore calling for EU rules that do not address only files that are already known.
What the IWF actually does
The IWF finds, assesses, and reports child sexual abuse imagery online. Confirmed files receive a digital fingerprint known as a hash. Platforms and law enforcement can use it to recognize and block known copies without opening the image again each time.
Generative systems make the problem harder: offenders can create new images or alter real photographs of victims. Each new file initially has no known fingerprint. It must be found and assessed before its hash can be added to blocklists. According to the IWF, these manipulations can retraumatize survivors even when the source image has been altered.
Why it matters
Of 6,221 images for which age and gender were recorded, 98 percent depicted girls. There were 2,534 images of children aged seven to ten, 2,369 of children aged eleven to 13, 1,004 of children aged three to six, and 190 of infants and toddlers under two. Children aged seven to 13 therefore accounted for 79 percent, up from 70 percent in 2025.
The Guardian notes that this statistic covers still images only. At the same time, the Report Remove service had already received 420 reports in 2026 from children concerning fake or manipulated intimate images of themselves, compared with 397 during all of 2025. This makes the debate practical: it concerns real victims, moderation, law enforcement, and how newly created content can be detected.
In plain language
A hash filter works like a list of previously stolen bicycles with unique frame numbers. If the same bicycle appears again, it can be recognized quickly. A generator, however, can keep producing new “frame numbers.” A list of known files is therefore not enough on its own; new cases must also be discovered, assessed, and reported.
A practical example
Suppose a platform receives 100,000 image uploads in one day. Hash matching can immediately filter out known prohibited files. But the remaining uploads may include newly generated variants that are not yet in any database. The platform therefore also needs additional, legally controlled detection, trained reviewers, secure reporting routes, and clear escalation rules.
For parents and young people, this means no single privacy setting provides a guarantee. Private accounts and limited recipient groups can reduce the public availability of personal photographs. They cannot replace platform safeguards or investigations when images are stolen or manipulated.
Scope and limits
First, the IWF measures only the material it assessed; the global total cannot be calculated from these figures. Second, the half-year results are not automatically comparable with every statistic from other hotlines because collection methods and laws can differ. Third, technology alone cannot solve the problem: automated detection can make mistakes, miss previously unseen content, and affect privacy and fundamental rights when deployed broadly.
The data also does not establish which model generated a particular image or how many offenders are behind the files. Effective rules must therefore address child protection, evidence preservation, independent oversight, and proportionate procedures together.
SEO & GEO keywords
Internet Watch Foundation, IWF, AI-generated child abuse, CSAM, hash matching, EU child protection, generative AI, platform moderation, online safety, deepfake abuse
💡 In plain English
The IWF found more illegal AI images in the first half of 2026 than during all of the previous year. Known images can be blocked by hash, but newly generated variants first have to be found and assessed.
Key Takeaways
- →The IWF assessed 6,310 AI-generated images as illegal in the first half of 2026.
- →That was 40 percent more than the 4,512 images recorded during all of 2025.
- →Girls appeared in 98 percent of images for which age and gender were recorded.
- →Hash matching blocks known files but does not automatically detect new variants.
- →The figures are IWF case data, not an estimate of the global total.
FAQ
What exactly did the IWF count?
It counted AI-generated still images reviewed between January and June 2026 that specialists assessed as child sexual abuse imagery under applicable law.
Why are hash blocklists not enough?
A hash recognizes known files. Newly generated or altered images initially have a different fingerprint and must first be discovered and assessed.
Are 6,310 images the worldwide total?
No. The number covers only images assessed by the IWF and cannot be used as a complete estimate for the internet.
Which age group appeared most often?
Children aged seven to 13 accounted for 79 percent of the images recorded by the IWF.