The AI-CSAM Explosion
A 26,385% surge in AI-generated child sexual abuse videos in a single year. How generative AI is fueling the worst child safety crisis of the digital age.
26,385%
increase in AI-generated CSAM videos
Source: IWF, 2025
15,036
total AI CSAM items found (2 years)
Source: IWF, 2025
485,000
NCMEC AI-CSAM reports (H1 2025)
Source: NCMEC, 2025
97%
of AI CSAM victims depicted as girls
Source: IWF, 2025
The Scale of the Crisis
The Internet Watch Foundation (IWF) documented a staggering 26,385% increase in AI-generated CSAM videos — from just 13 in 2024 to 3,443 in 2025. Across images and videos combined, IWF identified 15,036 pieces of AI-generated CSAM over two years.(Source: IWF Annual Report 2025)
AI-Generated CSAM Videos (IWF)
65% of AI-generated abuse videos were classified as Category A — the most severe — compared to 43% for non-AI material. AI doesn't just scale production; it escalates severity.(Source: IWF, 2025)
In a single forum, researchers found 20,254 AI-generated CSAM images posted in one month alone. 90% were realistic enough to be prosecuted under existing law.(Source: Stanford Internet Observatory / Cyberlaw)
How AI Tools Are Being Exploited
Offenders are exploiting generative AI in increasingly sophisticated ways. The IWF identified several key vectors:(Source: IWF, 2025)
LoRA Fine-Tuning
Offenders can fine-tune open-source image models using as few as 20 images of a real child in as little as 15 minutes, producing unlimited new abuse imagery of that specific victim.
Audio Deepfakes
AI voice cloning is being used to create audio of children in distress, adding another dimension to abuse material and enabling more convincing coercion.
Agentic AI
Autonomous AI agents are being deployed to automate CSAM distribution operations — searching for victims, managing accounts, and evading detection at scale.
The "Deepfake Defense"
Offenders are claiming that real abuse material is AI-generated to evade prosecution, creating a legal gray area that undermines investigations.
NCMEC Reporting Surge
Reports of AI-generated CSAM to the National Center for Missing & Exploited Children (NCMEC) have grown exponentially — from 4,700 in 2023 to 67,000 in 2024 (+1,325%) to 485,000 in just the first half of 2025.(Source: NCMEC, 2025)
NCMEC AI-CSAM Reports
Of the 485,000 H1 2025 reports, 380,000 came from Amazon AI Services alone — a single provider accounting for 78% of all AI-CSAM reports.(Source: NCMEC, 2025)
AI Platform Accountability
Major AI companies are now reporting CSAM generated through their platforms to NCMEC, but the numbers reveal how widespread the abuse of these tools has become:
OpenAI
32,079
+9,650% YoY
Source: NCMEC
Amazon AI
30,759
New reporter
Source: NCMEC
Anthropic
971
New reporter
Source: NCMEC
OpenAI reported 80x more CSAM in H1 2025 compared to H1 2024. While increased reporting is positive, it reflects the massive scale at which these tools are being abused.(Source: NCMEC, Fortune)
The "Deepfake Defense"
Perhaps the most insidious consequence of AI-generated CSAM is the "deepfake defense" — offenders caught with real abuse material are increasingly claiming it was AI-generated. This forces prosecutors to prove authenticity, slowing cases and creating reasonable doubt. As AI-generated imagery becomes indistinguishable from real photographs, this defense threatens to undermine prosecutions of actual abuse.(Source: IWF, 2025)
Commercialization & Weaponization
AI-generated CSAM is being commercialized for blackmail and sextortion schemes. Offenders create realistic fake images of a child, then threaten to distribute them unless the victim complies with demands — often for real abuse material, money, or further exploitation. The barrier to creating this material has collapsed: what once required access to real victims can now be generated from a school yearbook photo.(Source: Thorn, IWF)
Where CSAM Is Hosted
The infrastructure supporting CSAM distribution is concentrated in a handful of countries. The EU hosts 62% of known CSAM worldwide.(Source: IWF, 2025)
CSAM Hosting by Country
Legal Response
Lawmakers are beginning to respond, though legislation lags far behind the pace of AI development:
TAKE IT DOWN Act
Signed into law, May 2025Criminalizes non-consensual intimate images including AI-generated content; requires platforms to remove within 48 hours.
STOP CSAM Act
PendingWould strengthen platform liability and expand law enforcement tools for investigating AI-generated CSAM.
ENFORCE Act
PendingTargets the infrastructure enabling CSAM distribution, including hosting providers and payment processors.
UK Crime and Policing Bill
In progressWould criminalize AI tools specifically designed or modified to generate CSAM, including LoRA models trained on abuse imagery.
What Needs to Happen
1. Safety by design in AI models. Companies releasing generative AI must implement robust safeguards before deployment — not after abuse is discovered. This includes CSAM detection in training data, output filtering, and provenance tracking.
2. Criminalize AI CSAM generation tools. Possessing, distributing, or creating LoRA models or other tools specifically designed to generate CSAM should carry the same penalties as producing real abuse material.
3. Mandatory reporting with teeth. All AI platforms must report to NCMEC, with financial penalties for non-compliance and requirements for proactive detection — not just reactive reporting.
4. International hosting accountability. Countries hosting disproportionate amounts of CSAM — particularly the Netherlands and EU nations — must face diplomatic and economic pressure to act.
5. Close the deepfake defense loophole.Legislatures must explicitly state that AI-generated CSAM carries identical legal weight to real CSAM, eliminating the "it's not real" defense.
Sources
- Internet Watch Foundation (IWF) — Annual Report 2025
- National Center for Missing & Exploited Children (NCMEC) — 2023–2025 data
- Stanford Internet Observatory / Cyberlaw
- Fortune — AI platform NCMEC reporting data
- Thorn — Technology to defend children from sexual abuse