Childlight finds AI-related child sexual abuse material referrals topped 400,000 in 2025
The increase measures technology-company reports, not an equivalent rise in underlying abuse. Researchers say better detection may explain part of it.
Loading page…
The increase measures technology-company reports, not an equivalent rise in underlying abuse. Researchers say better detection may explain part of it.
Listen to this story
Childlight’s 2026 index frames the use of AI to fabricate sexual images of children as a measurement and protection-system problem, not just a content-removal challenge. It calls for reliable evidence to guide earlier intervention, survivor support, stronger technology-sector accountability and laws that keep pace with emerging threats. The institute also warns that referral data cannot establish the underlying prevalence of abuse, while population-level evidence remains thin across the regions it covers.
Technology companies referred more than 400,000 AI-related reports to NCMEC in 2025, up from about 4,700 in 2023; improved detection and reporting may explain part of the rise.
Only 22 of 103 countries and areas had representative survey data covering both offline and technology-facilitated abuse.
The index covers East Asia and the Pacific, North America, and Latin America and the Caribbean—regions home to about 63% of the world’s children.
Technology companies referred more than 400,000 reports of AI-related child sexual abuse material in 2025, up from about 4,700 in 2023, according to Childlight’s new research. Its 2026 Into the Light Index pairs that sharp increase with a warning: governments lack much of the evidence needed to understand where children face harm and prevent it.
The increase of more than 8,400% concerns reports submitted to the US-based National Center for Missing and Exploited Children. Childlight cautions that improved detection and reporting may explain some of the rise. The figures therefore track what companies identified and referred, rather than establishing an equivalent increase in the underlying abuse.
The material includes deepfakes made with so-called nudify apps—tools that create fabricated sexual images using a real child’s likeness. Researchers said these images can become instruments of blackmail, with children pressured to provide money or additional images. The fact that an image is fabricated does not remove its potential use against an identifiable child.
Debi Fry, Childlight’s global director of data, urged governments and technology companies to understand and prevent these emerging forms of abuse and remove material rapidly. Her response addresses both the production of abusive images and what happens after they appear: the institute is calling for prevention as well as faster removal.
We cannot effectively prevent abuse or protect children without understanding its true scale and nature.
Paul Stanfield, Childlight chief executive
The AI-related referrals are one part of a broader research project. The University of Edinburgh’s dataset record describes this third edition of the index as covering East Asia and the Pacific, North America, and Latin America and the Caribbean. Together, those regions account for approximately 63% of the world’s children. Its scope extends beyond online imagery to abuse offline and the systems intended to protect children.
The dataset brings together six areas: abuse-material metrics, online and offline victimisation, public-sector frontline records, child helpline data, and measures of protection-system capacity. Frontline data now reaches beyond policing into health, social care and education. A new system-strengthening category examines law-enforcement capabilities and the processes for turning data into action. Accompanying files explain variable definitions, sources and methodological details.
Yet basic gaps remain. Among 103 countries and areas examined across the three regions, only 22 had representative survey data covering both offline and technology-facilitated abuse, Childlight said. That finding concerns the availability of population-level evidence across both settings—not the number of countries submitting AI-related material reports. It is a separate weakness in the picture available to governments.
Childlight identified several groups as particularly underrepresented: children with disabilities, Indigenous children, migrants, children in humanitarian settings, very young children and those abused within their families. Fry said combining surveys, police records and other sources can help reveal risks, identify weaknesses in protection systems and direct resources toward the places where they are most needed.
The institute’s recommendations reach beyond detecting individual files. It wants governments to treat child sexual exploitation and abuse as a preventable public health emergency, investing in reliable measurement and using that evidence to support survivors and intervene earlier. It also calls for stronger technology-sector accountability, laws that keep pace with emerging threats, and prevention efforts beginning earlier in childhood.
The intended destination for better data is practical protection: stronger support for survivors and action before harm occurs. Childlight is asking governments to make that shift through sustained leadership, investment and international collaboration.
Loading discussion...
Join the conversation
Explain how better detection would affect your judgment.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.