IWF counted 6,310 AI-generated child sexual abuse images in six months

The Internet Watch Foundation (IWF) says its analysts assessed 6,310 AI-generated images that met the UK legal definition of child sexual abuse material between January 1 and June 30, 2026. That six-month count exceeds the total IWF published for the whole of 2025. Its October 5 announcement also calls for EU rules that allow services to detect previously unseen material as well as known files.

There is a discrepancy in the comparison year. The October announcement gives 4,512 images for 2025, while IWF's 2025 image report gives 4,586. The six-month 2026 count is higher than either figure. IWF has not explained the 74-image difference in those public documents, so an exact percentage increase is unsettled.

The big change

• What changed: In six months, IWF classified more AI-generated child sexual abuse image files than it recorded for all of 2025, whichever of its two published 2025 totals is used. The count measures its assessed material, not the amount circulating online.
• Why it matters: Each newly identified file needs assessment before its digital fingerprint can help services find exact copies. For child-safety teams, a rising intake makes the distinction between detecting known files and finding new ones consequential.
• What to watch: IWF wants the proposed EU Child Sexual Abuse Regulation to cover detection of previously unseen material. The scope of that proposal, and IWF's clarification of its 2025 baseline, determine how its warning can be assessed.

What the analysts counted

The October release breaks the 6,310 first-half images into UK severity categories: 5,557 in Category C, 403 in Category B and 350 in Category A. Category C accounted for about 88% of the assessed AI-generated images. Under IWF's published classification method, each image receives one severity assessment. Category C is the category for indecent images that fall outside Categories A and B. The categories describe legal severity; they do not tell us how many children or people created the files.

IWF's 2025 methodology says its data comes from public reports and analysts' proactive searches. It treats reports, webpages, images and videos as different units. For individual images, assessors record the age and sex of children depicted and create a digital fingerprint, or hash, after assessment. A composite file receives one severity grade, and even a small file change can produce a different hash. The 2025 image report cautions that children may appear repeatedly in the dataset. Image counts therefore cannot be read as counts of distinct children.

IWF's March 2026 research report says analysts recorded material as AI-generated only when its source or metadata indicated that status. That rule could leave some synthetic material outside the labelled count. The October announcement gives no number of images screened, no estimate of all material online, and no account of whether search effort or the AI labelling rule changed between the comparison periods. The figures show what IWF identified and assessed. They do not establish a population-wide rate or a cause for the increase.

Why the response distinguishes known and new files

Once IWF confirms an illegal image, it can share a hash with law enforcement and technology companies to help match known files. Its October announcement says this method cannot identify a new file before that file has been found and assessed. IWF uses that gap to argue that the proposed EU Child Sexual Abuse Regulation should permit detection of previously unseen material as well as matching known material. That is the organisation's policy position, not a measured result of the proposed law.

The new count gives policymakers and platform safety teams a concrete measure of IWF's assessed workload. Its unresolved 2025 baseline and limited coverage information matter when the figures are used to describe the scale of harm beyond the hotline's work.

Sources & further reading

• IWF announcement, October 5, 2026, for the January 1–June 30 count, category breakdown, stated 2025 comparison and IWF's EU policy request. Its 4,512-image 2025 baseline conflicts with the annual image report; the release does not explain the difference or provide a full first-half collection method.
• IWF 2025 image insights, for its 4,586 criminal AI-generated image total, separate prohibited-image figure and warning that the same child may appear in multiple images. This page is the conflicting first-party baseline, not a count of distinct children.
• IWF 2025 methodology, for how public reports and proactive searches enter the system, how images and reports differ, how severity and hashes are recorded, and the legal categories. It describes the 2025 reporting system; the October 2026 release does not publish a complete updated method.
• IWF, Harm without limits, March 2026, especially the hotline data section, for its stated rule that analysts record AI generation when indicated by the source or metadata. The report contains distressing descriptions; it does not provide the October first-half figures.