A report released by the UK’s Internet Watch Foundation on October 4 found that confirmed and removed AI-generated child sexual abuse material increased more than four-fold year-on-year over the past 12 months. The report notes that the spread of generative AI tools, combined with criminal underground markets, has dramatically lowered the barrier to producing and distributing such material, and calls on national legislatures to bring AI-generated CSAM under dedicated legal frameworks.

[Key Facts]

  • The reporting period covers July 2025 to June 2026; confirmed and removed AI-generated CSAM increased more than four-fold compared with the previous cycle.
  • The report indicates such content is largely produced using open-source image generation models, fine-tuned LoRA and low-barrier desktop tools.
  • Criminal underground markets have listed AI-generated CSAM as a standalone product category, forming complete production, distribution and money-laundering chains.
  • The report calls on national legislatures to clearly designate AI-generated CSAM as a separate offense, rather than simply subsuming it under existing “child sexual abuse material” statutes.
  • Multiple AI companies (including Stability AI, Midjourney and Runway) have committed to stricter generative content filtering mechanisms.

[What We Know]

  • Statistical scope: The report covers URLs submitted by UK-based and several English-language internet service providers, with a sample of over 500,000 URLs.
  • Reasons for growth: The report attributes the increase to three factors — capability improvements in open-source image generation models, shared fine-tuning data via custom LoRAs, and scaled criminal underground market operations.
  • Legal status: The UK’s current sexual abuse content legislation treats “real children” and “AI-generated simulations” under different clauses, causing conviction difficulties in some cases.
  • Industry response: Multiple AI companies have pledged to deploy stricter generative content filtering mechanisms and strengthen user-behavior auditing after the report’s release.
  • International cooperation: The report calls for a transnational database to share hash fingerprints and generative-tool signatures for AI-generated CSAM.

[Wider Perspective]

First, the negative spillover effects of generative AI have shifted from theoretical worry to actual data shock. Over the past two years, discussions of generative AI misuse have largely stayed at the “potential risk” level. This report’s measured four-fold year-on-year growth confirms that the speed and scale of AI technology misuse has exceeded most legislators’ and companies’ expectations. AI safety governance cannot stay at “principled” statements; it needs executable technical and legal tools.

Second, the dual-use dilemma of open-source models is especially acute in child-protection scenarios. After open-source image generation models like Stable Diffusion publicly released their weights, researchers struggle to limit their use for CSAM production. Academia and industry have attempted mitigation through content filtering, model watermarking and training-data auditing, but existing solutions have notable loopholes.

Third, the AI upgrade of criminal underground markets outpaces traditional law enforcement cadence. The report indicates criminals form complete industrial chains through cryptocurrency, anonymous networks and cross-border hosting; single-country enforcement is hard to address. INTERPOL, EUROPOL and national cybercrime units need to establish regular information sharing and joint enforcement mechanisms.

Fourth, the legislative “real vs. simulated” binary is poorly suited to the AI era. Current laws use “whether real children are involved” as the core distinction, but AI-generated simulated images cause no less psychological harm to children or damage to society than real content. The report urges legislators to bring “AI-generated simulations” and “real children” under the most stringent child-protection framework.

Fifth, AI companies need to embed child protection at the product-design phase. Product builders must integrate child-protection mechanisms during model training, data filtering, content generation and user-experience stages, rather than only passively responding after the fact. Whether the “pre-filtering” mechanisms pledged by Stability AI and others can land at the engineering level remains to be tested by 12 months of real cases.

[Viewpoints Compared]

  • Internet Watch Foundation CEO: Called it “one of the most worrying scenarios of AI technology misuse,” urging national legislatures to act immediately.
  • UK Home Secretary: In a parliamentary speech pledged to push legislation that clearly designates AI-generated CSAM as a separate offense.
  • Stability AI public statement said it has strengthened generative content filtering mechanisms, while conceding that “total blocking is nearly impossible.”
  • UNICEF representative: In a statement called on countries to establish a transnational database to share hash fingerprints for AI-generated CSAM.
  • Digital rights advocacy organization: In media interviews warned against expanding surveillance in the name of child protection, emphasizing technical solutions and human rights safeguards together.
  • U.S. National Center for Missing and Exploited Children: Said the U.S. is working with British, Canadian and Australian counterparts to push for a North America-and-overseas sharing mechanism.