IWF says AI child abuse image detections rose 40% in 2026, intensifying pressure on EU lawmakers

The Internet Watch Foundation says its analysts identified 6,310 AI-generated child sexual abuse images in the first six months of 2026, more than the 4,512 it recorded across all of 2025. That is a 40% increase on last year’s total, and it underlines the scale of IWF’s identified cases, although the tally does not measure every image circulating online. The IWF published the data on 5 October 2026 and used it to press EU policymakers to finish long-delayed child protection legislation. IWF report

The most important detail is not just the volume. It is the shape of the content. The IWF says that, among the cases where age and gender were recorded, girls accounted for 98% of the imagery in the first half of 2026. It also says the share depicting children aged 7 to 13 was 79%, up from 70% across 2025. The data further include images showing infants and toddlers under two. In other words, more of the identified images depict younger children, according to the IWF’s classification. The figures describe depicted ages, not a count of identifiable real-world victims.

That matters because the IWF’s own remediation model depends on identification. The organization converts confirmed criminal images into hashes, or digital fingerprints, so platforms and law enforcement can block repeats automatically without keeping the original file. That works well against content that has already been found. It does not work against newly generated material that has never been seen before, which is why the IWF argues that regulations limited to known content will leave a large gap open. The technical problem is simple to state and difficult to solve: if abusive AI output can be produced at scale faster than it can be reviewed and fingerprinted, purely reactive moderation will always be behind.

This is why the EU angle is not just political theater. The IWF says the proposed Child Sexual Abuse Regulation should give services legal certainty to detect and remove both known and unknown material. That is a practical boundary, not an abstract one. A law that only protects hash-based blocking helps with repeat distribution. A law that also allows proportionate detection of previously unseen abusive content changes the operational options available to platforms, especially those that currently hesitate because of legal uncertainty around scanning and removal. The organization’s argument is that the risk is no longer limited to re-sharing old abuse images; AI lowers the cost of producing fresh ones.

The wider reporting context reinforces that concern. The Guardian’s account of the same IWF data noted that the figures refer to still images, even as AI-generated abuse videos have also surged in recent years. It also reported that the UK government has already moved to criminalize adapting AI models to create abuse material or distributing such models, showing that at least one European policymaker has already chosen a harder line on model misuse. Guardian report

A separate IWF breakdown helps explain what sits behind the headline total. Its analysts classified 5,557 of the first-half 2026 images as Category C under UK law, or 88% of the AI images identified in that period, compared with 2,842 and 62% across all of 2025. The IWF also recorded cases in the more severe categories. These are classifications of material reviewed by one hotline, not a population-wide severity survey. The shift matters for moderation design because systems must identify a range of illegal material; treating the entire problem as one narrow category would miss much of what the analysts actually found.

The Guardian reports a related but distinct signal from Report Remove: 420 children had reported images of themselves that they believed were faked or manipulated to appear explicit, already above the 2025 total of 397. Those reports reflect what children believed about their own images; they are not the same dataset as the IWF’s confirmed illegal AI-generated images and must not be added to the 6,310 count. They do, however, show why the impact is more than a content-moderation metric. Manipulated images can involve the likeness of real children, and a response needs reporting and removal paths that children can actually use.

For safety teams, trust-and-safety leads and lawmakers, the operational takeaway is narrow but important. The IWF’s figures support three separate decisions: first, child-safety systems cannot treat AI abuse as a niche edge case; second, hash-matching remains necessary but is no longer sufficient on its own; and third, the legal question now is whether platforms may be allowed, and required, to detect material that has never been hashed before. The data do not prove the total size of the problem across the internet. They show why each newly generated image must first be found and assessed before hash-matching can block copies of it.

That limitation matters. The IWF’s counts describe what its analysts identified, not the full online universe, and the legal categories are tied to UK law. Even so, the identified count has risen, and the age distribution has shifted toward younger depicted children. The IWF argues that newly generated material exposes a gap in a response built only around previously known images. For EU lawmakers, the question is no longer whether AI can be abused this way; it is whether the next regulation will be written to catch only yesterday’s material or to make room for action against the next batch before it spreads. Watch whether the EU Child Sexual Abuse Regulation permits proportionate detection of previously unseen material, because that will shape how platforms can respond to new AI-generated abuse.