Analysts at the Internet Watch Foundation found 6,310 AI-generated child sexual abuse images in the first six months of 2026, 40% more than the whole of 2025. The UK's Crime and Policing Act 2026 has created a new offence, carrying up to five years in prison, for making or adapting an AI model built to produce such imagery.
AI Child Abuse Imagery Up 40% This Year
The figures, published on Monday, 5 October 2026, mark a sharp rise on the 4,512 images recorded across all of 2025. Girls accounted for 98% of the images where age and gender were recorded.
The IWF's annual report found that 65% of the 3,443 AI-generated abuse videos identified in 2025 fell into Category A, the most severe legal classification, compared with 43% of non-AI videos. Children under two appeared in 190 of the images recorded so far this year.
The charity's Report Remove service has also received 420 reports this year from children who believed their images had been faked into explicit content, whether or not that material ultimately met the criminal threshold, already above last year's total of 397.
The Law's New Reach
Section 72 of the Act inserts a new offence into the Sexual Offences Act 2003 covering a 'CSA image-generator,' defined broadly as a thing, which can mean a program, a service or information in electronic form, made or adapted to create or facilitate the creation of child sexual abuse images.
Making, adapting, possessing, supplying or offering to supply one carries up to five years' imprisonment on indictment. Equivalent provisions apply separately in Scotland and Northern Ireland.
Home Office minister Natalie Fleet told Parliament the measure targets AI models that produce 'hyper-realistic' abuse imagery that 'often contains the likeness of real children.'
As of this week, that offence appears to remain uncommenced, according to legislation.gov.uk's own classification of the section as prospective. Making, possessing or distributing AI-generated CSAM is already a criminal offence under existing UK law regardless. What the new offence adds is liability for the tools and models built specifically to produce it.
A separate new offence under section 77 targets those who moderate or administer websites hosting such material, carrying up to 10 years in prison, and appears to share the same uncommenced status.
The Act also expands the existing possession-of-advice offence under the Serious Crime Act 2015 to cover guidance on producing pseudo-photographs and prohibited images, closing a gap that previously excluded AI-generated material from the 'paedophile manual' offence.
A related offence, covering tools that generate fake nude or 'nudification' images of real adults, carries a maximum of three years and has been in force since 29 June 2026. It does not apply to images of children, since the law treats indecent images of children, deepfakes included, as already covered elsewhere.
Where Enforcement Breaks Down
Meta, the parent company of Facebook and Instagram, ran more than 300 paid ads containing AI-generated child sexual abuse imagery between November 2025 and August 2026, according to the Tech Transparency Project, whose findings were reported exclusively by WIRED.
An initial batch of roughly 50 ads was reported in August; a follow-up investigation in September found more than 250 further ads, some using images of real children, including a member of a European royal family.
Meta said its automated systems review ads before publication and that it removed 36 million pieces of child exploitation content last year, though researchers found new offending ads appearing even after the company was contacted about the first batch.
San Francisco's city attorney has since sent Meta a cease-and-desist letter demanding an explanation for how the ads kept running. Apple separately removed an app called MaskAI from its App Store for breaching its ban on nudification tools.
The Pushback
The 8th US Circuit Court of Appeals this month granted an injunction suspending enforcement of Minnesota's new ban on AI 'nudification' tools, while Elon Musk's AI company, xAI, pursues a constitutional free-speech challenge to the law. A lower court had already rejected xAI's bid for the injunction, ruling the company had waited too long to seek emergency relief.
xAI says its Grok Imagine tool carries 'rigorous protections' against generating sexualised images of real people without consent, and has separately sued users accused of bypassing those safeguards.
The EU has separately been advancing its own rules in this area, agreeing in principle to a criminal-law framework that would criminalise designing or adapting AI systems to produce CSAM.
That is alongside reported moves to restrict AI tools that generate non-consensual sexual imagery more broadly, though the exact scope and timing of each instrument is worth checking against the Council's own text before publication.
The picture across both jurisdictions is one of law reaching further into how these tools are built than enforcement, so far, has managed to close, whether that's offences still awaiting commencement in the UK or a US court weighing a constitutional challenge before Minnesota's law can take effect.
