SafetyRisk, alignment & guardrails
AI-related child sexual abuse reports to NCMEC rose more than 8,400% in two years, Childlight data shows
AI-related child sexual abuse reports to NCMEC rose from about 4,700 in 2023 to over 400,000 in 2025, according to Childlight. Most reports involve indirect or unclear AI links, while separate IWF data shows real growth in AI-generated imagery.
From the The Guardian · AI article of November 16, 2026 and cited sources: Childlight, a child safety institute at the University of Edinburgh, based its 8,400%-plus figure on NCMEC's yearly totals of AI-related reports: about 4,700 in 2023, 67,000 in 2024 and more than 400,000 in 2025. [1] [2] [7]
SOURCES[1] Reports of AI-related child sexual abuse material rise… theguardian.com[2] NCMEC: Generative AI missingkids.org[7] Childlight: Into the Light Index childlight.org
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- Childlight, a child safety institute at the University of Edinburgh, based its 8,400%-plus figure on NCMEC's yearly totals of AI-related reports: about 4,700 in 2023, 67,000 in 2024 and more than 400,000 in 2025. [1] [2] [7]
- Original source
- The Guardian · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 16, 2026 article published
- Covered
- Nov 16 edition
- Evidence
- 10 sources across 9 sites
- 1Reports of AI-related child sexual abuse material rise…theguardian.com
- 2NCMEC: Generative AImissingkids.org
- 3Bloomberg: Amazon found child sex abuse in AI training databloomberg.com
- 4Stanford CIS: Letter to NCMEC about AI-CSAM report…cyberlaw.stanford.edu
- 5WCAX: Reports of AI-generated child sexual abuse material…wcax.com
- 6Free Press Journal: AI-generated CSAM images surge 40%freepressjournal.in
StrongHow it’s measured
Hacker News · 17 likes · 15 comments
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.

Childlight, a child safety institute at the University of Edinburgh, based its 8,400%-plus figure on NCMEC's yearly totals of AI-related reports: about 4,700 in 2023, 67,000 in 2024 and more than 400,000 in 2025. The figure counts reports that companies sent to NCMEC. It does not count abuse images. [1] [2] [7]
Most of the 2025 reports are not about AI-generated abuse imagery. More than 200,000 had an AI connection but no clear use, and more than 145,000 involved altering existing abuse files. About 7,000 involved generating or possessing AI-made material. Amazon's large reporting volume came from matching known real imagery while scanning training data. [2] [3] [4]
The Internet Watch Foundation's own analyst assessments show clear growth in AI-generated material. It found 6,310 images in the first half of 2026, compared with about 4,500 in all of 2025, and 1,286 videos, up from two in the same period a year earlier. 65% of 2025's AI videos were rated in the most severe category. [6] [8]
Labelling is inconsistent. Companies flagged about 11,000 items as AI-generated, while NCMEC staff identified more than 158,000. Reports without data provenance are often unusable, so teams should scan training datasets, record where data came from, and add safeguards to file-upload and image-editing features. [2] [3]
The report counts are documented, but they measure reporting volume, not the amount of abuse imagery.
Read the full assessment
IWF analyst data independently shows AI-made images and video increasing, which puts pressure on generation services to label outputs, screen training data and guard upload and editing features.
Executive brief
The 8,400% rise in "AI-related" child sexual abuse reports is real in the sense that the counts went up: reports to the US National Center for Missing and Exploited Children (NCMEC) grew from about 4,700 in 2023 to more than 400,000 in 2025 (Guardian). NCMEC's own breakdown puts more than 200,000 of the 2025 reports in an "AI connection, use unclear" category. Separate figures from the Internet Watch Foundation (IWF), a UK monitoring charity, do point to real growth in AI-made imagery.
Read the full section
The 8,400% rise in "AI-related" child sexual abuse reports is real in the sense that the counts went up: reports to the US National Center for Missing and Exploited Children (NCMEC) grew from about 4,700 in 2023 to more than 400,000 in 2025 (Guardian). Most of those reports, however, are not reports of AI-generated abuse imagery. NCMEC's own breakdown puts more than 200,000 of the 2025 reports in an "AI connection, use unclear" category. Only about 7,000 involve generating or possessing AI-made abuse material (NCMEC). Separate figures from the Internet Watch Foundation (IWF), a UK monitoring charity, do point to real growth in AI-made imagery. The headline percentage measures reporting volume, not the amount of abuse material.
What changed and event timeline
Amazon's reports came from training-data scans
Bloomberg reported that Amazon's AI-related reports to NCMEC were matches against databases of known real abuse imagery, found while scanning training data, and that none were AI-generated (;).
Stanford researcher asks NCMEC to redesign its form
Riana Pfefferkorn argued that NCMEC's single "Generative AI" checkbox mixes up very different situations, and that this led to months of misleading "AI CSAM crisis" coverage ().
NCMEC describes the surge
NCMEC cited 4,700 reports in 2023 and 67,000 in 2024. The article says the 400,000-plus figure covered only the first half of 2025, which conflicts with full-year framing elsewhere ().
IWF's half-year count passes all of 2025
The IWF assessed 6,310 AI-generated abuse images in the first half of 2026, compared with about 4,500 in the whole of 2025. It also found 1,286 AI videos, up from two in the same period a year earlier ().
Childlight publishes the 8,400% figure
Childlight, a child safety institute at the University of Edinburgh, released the comparison in its regional report covering North America, East Asia and the Pacific, and Latin America and the Caribbean (;).
Capabilities and access
- None of the reviewed sources names a specific AI model or version.
- The tools linked to the harm are "nudify" apps, which put a child's face onto explicit imagery, and general image and video generators (Guardian).
- NCMEC's largest specific category is people using AI to alter existing abuse files: more than 145,000 reports in 2025.
Read the full section
- None of the reviewed sources names a specific AI model or version.
- The tools linked to the harm are "nudify" apps, which put a child's face onto explicit imagery, and general image and video generators (Guardian).
- NCMEC's largest specific category is people using AI to alter existing abuse files: more than 145,000 reports in 2025. There were also more than 30,000 reports of attempted generation by uploading a file (NCMEC).
- The IWF says AI-generated video has improved sharply. It recorded 3,443 AI videos in 2025, and 65% fell into Category A, the most severe grade (IWF AI report).
Technical analysis for researchers and developers
- Childlight's figure is a count of reports tech companies sent to NCMEC. It does not count images (Guardian).
- How NCMEC breaks down 2025:
- more than 12,000 reports of abuse material found in training data
Read the full section
- What the metric counts. Childlight's figure is a count of reports tech companies sent to NCMEC. It does not count images (Guardian).
- How NCMEC breaks down 2025:
- more than 12,000 reports of abuse material found in training data
- more than 7,000 of generating or possessing AI-made abuse material
- more than 30,000 of attempted generation by file upload
- more than 145,000 of manipulating existing abuse files
- more than 3,000 of chat-based exploitation
- more than 200,000 with an AI connection but unclear use
(NCMEC)
- Labelling gap. Companies have been able to flag content as AI-generated since October 2023, yet they labelled only about 11,000 items that way. NCMEC staff identified more than 158,000 (NCMEC).
- Hash matching. Hash matching compares files against a database of known images. It cannot detect new synthetic material, which is how Amazon's training-data scans produced large counts of known real imagery (Stanford CIS).
Claims and evidence
- 4,700 to 400,000-plus reports. These match NCMEC's published yearly figures (NCMEC). Childlight did not run its own detection.
- The Guardian says "more than 8,400%"; Childlight's index page says 8,411% (Childlight).
- The 2025 total is contested.
Read the full section
- 4,700 to 400,000-plus reports. These match NCMEC's published yearly figures (NCMEC). Childlight did not run its own detection.
- The headline percentage. The Guardian says "more than 8,400%"; Childlight's index page says 8,411% (Childlight).
- The 2025 total is contested.
- Pfefferkorn cites 485,000 AI-related reports in the first half of 2025 alone, of which 380,000 came from Amazon (Stanford CIS).
- Bloomberg describes more than one million AI-related reports in 2025, mostly from Amazon (Bloomberg).
- The reviewed sources don't explain how NCMEC got from those counts to its 400,000-plus total.
- IWF figures. These come from its analysts' own assessments, so they are a separate data source from the NCMEC counts (IWF).
Context and prior work
- Childlight's Into the Light Index has covered Western Europe and South Asia before.
- Across all categories, NCMEC received 21.3 million CyberTipline reports in 2025 (NCMEC data).
- The IWF has tracked AI-generated abuse material since analysts first spotted it in early 2023 (IWF).
Read the full section
- Childlight's Into the Light Index has covered Western Europe and South Asia before. Its 2026 data update, launched at the World Health Assembly, estimated that 26.6% of survey respondents experienced online sexual solicitation before age 18 (EdTech Innovation Hub; Childlight).
- Across all categories, NCMEC received 21.3 million CyberTipline reports in 2025 (NCMEC data).
- The IWF has tracked AI-generated abuse material since analysts first spotted it in early 2023 (IWF).
Limitations, safety and contested findings
- Childlight's Prof Debi Fry acknowledges that better detection and reporting may explain part of the rise (Guardian).
- NCMEC itself says that not every AI-related report involves creating or sharing AI-generated abuse material (NCMEC).
- Sources disagree on whether the 400,000-plus figure covers the first half of 2025 or the whole year (WCAX).
Read the full section
- Childlight's Prof Debi Fry acknowledges that better detection and reporting may explain part of the rise (Guardian).
- NCMEC itself says that not every AI-related report involves creating or sharing AI-generated abuse material (NCMEC).
- Sources disagree on whether the 400,000-plus figure covers the first half of 2025 or the whole year (WCAX).
- No source independently corroborates Childlight's regional estimates beyond the NCMEC data it draws on.
- The harm itself is not disputed. The IWF reports that 97–98% of AI-generated abuse imagery depicts girls, and that more and more children report fake or manipulated sexual images of themselves (IWF; Free Press Journal).
Business and practitioner implications
- Teams building models should hash-scan their datasets, keep records of where data came from, and pass that origin information on when they report.
- Image and video generation services should use NCMEC's AI-generated flag consistently, which narrows the 11,000-versus-158,000 gap (NCMEC).
- File-upload and image-editing features need controls against misuse, since altering existing files and upload-based generation are the largest specific categories (NCMEC).
Read the full section
- Training data. Teams building models should hash-scan their datasets, keep records of where data came from, and pass that origin information on when they report. NCMEC says reports without it are often unusable (Bloomberg).
- Labelling. Image and video generation services should use NCMEC's AI-generated flag consistently, which narrows the 11,000-versus-158,000 gap (NCMEC).
- Safeguards. File-upload and image-editing features need controls against misuse, since altering existing files and upload-based generation are the largest specific categories (NCMEC).
- Policy pressure. Childlight is urging governments to treat child sexual exploitation as a preventable public health emergency (Guardian).
The source trail.
Sources (10)
Reports of AI-related child sexual abuse material rise 8,400% in two years, data shows
Article text retrieved; extracted text may omit tables or interactive elements.
theguardian.com