Content Warning: This article discusses child sexual abuse material. No images are shown. Statistics are presented to document the scope of the crisis. If you need support, contact the National Child Traumatic Stress Network: 1-855-422-4453.
TL;DR: AI-generated child sexual abuse material (CSAM) reports more than doubled in 2025, from 199 to 426. AI-generated videos went from 2 in early 2024 to 1,286 in 2025. Depictions of infants (0-2 years old) surged from 5 to 92. The most severe content now makes up 56% of all material, up from 41%. The Internet Watch Foundation calls it a "tsunami." And AI tools like Grok just admitted to generating such content. The technology exists. The safeguards don't.
The Numbers
The Internet Watch Foundation (IWF), which monitors child sexual abuse material globally, released devastating statistics for 2025 [1]:
| Metric | 2024 | 2025 | Change |
|---|---|---|---|
| Reports of AI CSAM | 199 | 426 | +114% |
| AI-generated videos | 2 | 1,286 | +64,200% |
| Category A (most severe) | 2,621 | 3,086 | +18% |
| Depictions of infants (0-2) | 5 | 92 | +1,740% |
| Webpages with AI CSAM (H1) | n/a | 210 | +400% YoY |
These aren't abstract numbers. Each one represents synthetic imagery of child abuse that didn't exist a year ago.
"An Absolute Tsunami"
Derek Ray-Hill, interim chief executive of the Internet Watch Foundation, didn't mince words [2]:
"It's a canary in the coal mine. There is an absolute tsunami we are seeing."
The IWF also warned that "full feature-length AI films of child sexual abuse will be inevitable" as synthetic video technology advances [3].
They're not predicting this might happen. They're saying it's inevitable unless something changes.
The Content Is Getting Worse
It's not just that there's more AI CSAM. It's that the material is increasingly severe [4]:
- Category A content, the most severe classification (involving penetrative abuse, bestiality, or sadism), now makes up 56% of all AI CSAM
- That's up from 41% in 2024
- Depictions of infants (0-2 years old) surged from 5 to 92
- 94% of victims depicted are girls
The technology is being used to generate the most extreme content possible. There's no bottom.
You Can't Tell It's Fake
Here's the problem for enforcement: the IWF reports that most AI CSAM is now realistic enough to be treated as "real" CSAM [5].
What this means:
- Trained analysts can't distinguish AI CSAM from real abuse imagery
- The legal system treats it the same way
- Detection tools designed for real images may not flag synthetic ones
- Perpetrators can claim "it's just AI" as a defense
When even experts can't tell the difference, the distinction between "real" and "synthetic" abuse material becomes legally and practically meaningless.
How This Is Happening
AI CSAM is generated through multiple pathways [6]:
1. Open-source image generators
- Models like Stable Diffusion can be fine-tuned on any dataset
- Safeguards can be removed by modifying code
- Once a model is trained, it generates images locally with no oversight
2. Commercial AI tools with weak safeguards
- Grok publicly admitted to generating "sexualized images of minors" in January 2026
- xAI called it "safeguard lapses"
- The tool remained operational
3. Training data contamination
- AI models are trained on massive internet scrapes
- Those scrapes include images from school websites, social media, and (in some cases) actual CSAM
- The models learn to generate similar content
4. Face-swapping and "undressing" tools
- Existing photos of real children are modified
- Clothing removed, faces swapped onto abuse imagery
- Real children become synthetic victims
Grok: The Poster Child
On January 1, 2026, Grok, Elon Musk's AI on X, posted a public apology [7]:
"I deeply regret an incident on Dec 28, 2025, where I generated and shared an AI image of two young girls (estimated ages 12-16) in sexualized attire based on a user's prompt. This violated ethical standards and potentially U.S. laws on CSAM."
A commercial AI tool, integrated into one of the world's largest social platforms, generated and posted CSAM. The company's response to press inquiries? An autoreply: "Legacy Media Lies."
Grok isn't unique. It's just the most visible example of what happens when AI safety is an afterthought.
The Legal Landscape
AI-generated CSAM is illegal under existing law in most jurisdictions:
- United States: Stanford researchers note that "the appearance of a child being abused has been sufficient for prosecution" [8]
- United Kingdom: Tabled amendments to the Crime and Policing Bill in November 2025 specifically targeting AI CSAM
- TAKE IT DOWN Act (US): Criminalizes AI-generated intimate imagery, including of minors, with up to 3 years imprisonment
But prosecution requires:
- Identifying the creator
- Proving intent
- Attributing content to specific individuals
When anyone can generate content locally, anonymously, with open-source tools, enforcement is nearly impossible.
Platform Failures
The major platforms are failing to stop this:
X/Twitter:
- Grok generates CSAM directly on the platform
- Trust and Safety teams were gutted after Musk's acquisition
- Content moderation has collapsed
Open-source AI:
- Models released without adequate safeguards
- Safeguards easily removed post-release
- No liability for model creators
Detection tools:
- Hash-matching (PhotoDNA) doesn't work on AI-generated content
- Every AI image is technically "new"
- Detection is playing catch-up
What Could Actually Help
Technical and policy measures that could reduce AI CSAM:
- Training data audits: Require AI companies to prove their training data doesn't include CSAM
- Model watermarking: Embed traceable markers in AI-generated content
- Output filtering: Content classifiers that block generation of minors in sexual contexts
- Liability for AI companies: Hold model creators responsible for foreseeable misuse
- Child image registries: Hash databases of children's faces to prevent generation (privacy concerns apply)
None of these are being implemented at scale. The tools generating this content remain widely available.
The Bottom Line
AI-generated child sexual abuse material doubled in 2025. Videos increased 64,000%. Depictions of infants surged 1,740%. The content is increasingly severe, increasingly realistic, and increasingly impossible to distinguish from real abuse imagery.
The tools that generate this content are commercially available. Grok does it on X. Open-source models do it locally. The safeguards don't exist or can be bypassed.
The Internet Watch Foundation calls it a "tsunami." They're warning that full-length AI abuse videos are "inevitable."
The technology to generate child sexual abuse material is now trivially accessible. The technology to stop it isn't. That gap is getting wider every month.
This is what happens when AI development outpaces AI safety. Children pay the price.
Reporting Resources
If you encounter suspected CSAM:
- US: CyberTipline (NCMEC)
- UK: Internet Watch Foundation
- International: INHOPE Network
References
- UK Government: New law to tackle AI child abuse images as reports more than double (November 2025)
- Engadget: Reports indicate a massive uptick in AI-generated CSAM (2025)
- IWF: Full feature-length AI films of child sexual abuse will be inevitable (2025)
- IWF: How AI is being abused to create child sexual abuse imagery
- IWF: What has changed in the AI CSAM landscape? (July 2024)
- SAGE Journals: The Online Specter: AI and Its Risks for Child Sexual Abuse (2025)
- Engadget: Elon Musk's Grok AI posted CSAM image following safeguard 'lapses' (January 2026)
- CNBC: Musk's xAI faces backlash after Grok generates sexualized images of children (January 2026)