The Verified Number · N°19

AI-linked child abuse reports are up 85-fold in two years

This figure tells two stories.

5 min read
AI-linked child abuse reports are up 85-fold in two years

A figure multiplied by 85

4,700 in 2023. More than 400,000 in 2025. These are reports of child sexual abuse material linked to artificial intelligence, multiplied by 85 in two years: an increase of more than 8,400%, announced on November 16 by Childlight, a child protection research institute attached to the University of Edinburgh. The Guardian ran it the same day, in its headline.

Before going further, two things. The phenomenon is real, and serious: we come back to it at the end, with numbers. And the math is right: 400,000 divided by 4,700 is 85.

What we checked is what that multiplier actually measures. To find out, you have to go back to the counter.

Who counts, and how

The figures don't come from Childlight. They come from NCMEC, the US national center for missing and exploited children. Its platform, the CyberTipline, receives reports of child sexual exploitation. US-based tech companies are legally required to report what they find on their services, and the public can report too. In 2025 it received 21.3 million reports, for all reasons combined.

Since January 2023, the center has tracked reports with a link to generative AI separately. The 400,000 make up about 2% of the total.

A counter like this works like a rain gauge under a downspout: it measures what falls into it. Put a second downspout over it and it overflows, without any extra rain.

What 2023 counted

2023 is the first year of tracking. The checkbox that lets companies flag an image or video as AI-generated was only added to the form in October of that year.

More to the point, the AI tools themselves were reporting almost nothing. NCMEC said so in its annual report: more than 70% of AI-related reports came from conventional platforms, which, in its view, indicated that most AI platforms, where content is created, weren't reporting.

The official company-by-company lists bear this out. In 2023, OpenAI sent NCMEC 329 reports, Anthropic one, and xAI, Elon Musk's company, doesn't appear. In 2025, the three sent 339,128 between them: 182,844 for OpenAI, 135,373 for xAI, 20,911 for Anthropic.

One caveat matters here. Those totals cover every reason for reporting, and NCMEC doesn't publish how many carry the AI flag, company by company. So nobody can say what share of the 400,000 comes from them. What can be said is that in two years, the generators entered the reporting system.

What 2025 counts

NCMEC breaks its 2025 total of more than 400,000 reports into five categories, plus a remainder.

  • More than 7,000 involve users who generate or possess AI-produced child sexual abuse material.
  • More than 30,000 involve attempts to generate it.
  • More than 145,000 involve the use of an AI tool on an existing file.
  • More than 12,000 involve material that companies say they found in training data: pre-existing content, in other words, not generated content.
  • More than 3,000 involve other uses.
  • And more than 200,000, half of the total, flag a link to AI without enough information to tell how AI is involved.

NCMEC says it itself: not every AI-related report involves creating or sharing generated content. The word "related" is a catch-all bag, holding a fabricated image just as easily as an existing file run through a tool, or a report about which nothing is known.

The figure that shrank

That 2025 figure has already changed once. Until mid-July 2026, NCMEC's page on AI announced 1.5 million reports for 2025, noting that 1.1 million came from Amazon's AI arm and contained no actionable information. Between July 15 and 27, according to archived versions of the page, the total became "more than 400,000". According to Bloomberg, which investigated in January, Amazon said the reports were matches against already-known material, found in the training data of its models, and that none of it had been generated by AI.

NCMEC's half-year figure from September 2025 tells the same story in its own way: 440,419 AI-related reports in the first six months of 2025. That's more than the full year as it's counted today, because the half-year figure still includes Amazon.

Childlight's report, for its part, carries both versions. Its summary mentions 1.5 million reports in 2025. More than 150 pages later, it settles on 400,000 and explains that Amazon's submissions could easily be mistaken for an explosion of AI-generated content, when they reflected an unusual reporting practice.

Where the reading slips

The +8,400% isn't in the report. It's in Childlight's press release, which attributes it to reports of AI-related child sexual abuse material. The release adds a note of caution: according to the researchers, changes in detection and reporting may contribute to the increase.

The Guardian faithfully repeats the figure in its headline. The slip happens in the first sentence of the article, which describes reports of AI being used to sexually exploit children. The caution only shows up near the end, paraphrasing Debi Fry, Childlight's director of data: better detection could explain part of the increase. Neither the makeup of the 400,000, nor the revised figure, nor Amazon's role is mentioned.

Nobody lied. A press release set two years side by side that weren't counted the same way, and the sentence saying so stayed at the bottom of the page.

What stays true

The underlying problem is serious, and NCMEC documents it. It has identified more than 275 direct victims of AI-generated content, and in many cases the perpetrator was someone close to the child. The center's analysts classified more than 158,000 images and videos as AI-generated between 2023 and 2025. The rise is real.

Part of the +8,400% is even good news: companies that reported nothing started reporting. A report is, first of all, a detection.

But the figure that should worry us may not be the one that made the headline. Over the same period, companies flagged only a little over 11,000 images and videos as AI-generated. The center itself found 158,000.

Was this article useful?

Topics covered:

SecurityAnalysis

Frequently asked questions

Where does the +8,400% figure come from?
From a press release by Childlight, a child protection research institute attached to the University of Edinburgh, presented on November 16, 2026. It compares 4,700 reports of AI-related child sexual abuse material in 2023 with more than 400,000 in 2025, all received by NCMEC, the US national center for missing and exploited children. The math is right: that is a multiplication by 85. The +8,400% is not in Childlight's report, only in its press release.
Why are there so few reports in 2023?
It was the first year of tracking, and the checkbox that lets companies flag AI-generated content was only added to the reporting form in October 2023. More importantly, according to NCMEC, most AI platforms weren't reporting: more than 70% of AI-related reports came from conventional platforms. In 2023, OpenAI sent 329 reports, Anthropic one, and xAI doesn't appear; in 2025 the three sent 339,128, for all reasons combined. NCMEC doesn't publish how many carry the AI flag company by company, so nobody can say what share of the 400,000 comes from them.
Are the 400,000 reports about AI-generated content?
Not all of them, far from it: AI-related does not mean AI-generated. NCMEC itself says that not every AI-related report involves creating or sharing generated content. More than 145,000 involve the use of an AI tool on an existing file, more than 12,000 involve pre-existing material that companies say they found in training data, and more than 200,000, half of the total, flag a link to AI without enough information to tell how AI is involved.
Why do some documents mention 1.5 million reports for 2025?
Until mid-July 2026, NCMEC's page on AI announced 1.5 million reports for 2025, noting that 1.1 million came from Amazon's AI arm and contained no actionable information. Between July 15 and 27, according to archived versions of the page, the total became "more than 400,000". Childlight's report carries both versions: 1.5 million in its summary, 400,000 further on.
Did the Guardian distort the figure?
Not in its headline, which faithfully repeats the figure from the press release. The slip happens in the first sentence of the article, which talks about reports of AI being used to sexually exploit children. Childlight's caution about detection and reporting only comes near the end, and neither the makeup of the 400,000, nor the revised figure, nor Amazon's role is mentioned.
Is the rise in AI-related child abuse material real?
Yes, the phenomenon is real. NCMEC has identified more than 275 direct victims of AI-generated content, and in many cases the perpetrator was someone close to the child. Its analysts classified more than 158,000 images and videos as AI-generated between 2023 and 2025, while companies had flagged only a little over 11,000 as such. Part of the +8,400% also comes from companies that reported nothing and started reporting.
Katja Liersch

Katja Liersch

Co-founder & Journalist

Katja is a journalist and TV producer. With decades of experience in mainstream media, she brings to Declic Media the perspective of those discovering AI: curious, demanding and pragmatic. She ensures every piece of content truly speaks to everyone.

All articles by Katja →
The free AI newsletter