AI-generated CSAM is overwhelming victim identification

Investigators can no longer assume that images, videos, or audio of abuse involve a real child. Online forums and seized devices now hold AI-generated child sexual abuse material (CSAM) alongside documented abuse of real children. Offenders can produce new material at almost unlimited volume, enough to bury the real children investigators need to find. Time spent trying to identify a child who doesn’t exist is time taken from a child who needs to be found.

The strain on the investigative process

The Internet Watch Foundation (IWF) found 3,443 AI-generated child sexual abuse videos in 2025. In 2024, it found 13 (IWF). Its analysts say much of the imagery now looks indistinguishable from photographic abuse. Some AI videos place known victims into new abuse scenarios and harm those survivors a second time.

The U.S. numbers need more care. The National Center for Missing & Exploited Children (NCMEC) logged more than 1.5 million CyberTipline reports with a generative AI link in 2025. But 1.1 million came from a single company and held nothing actionable (NCMEC). The number investigators should watch is smaller, at more than 145,000 reports of people using AI tools to alter existing abuse files. Each one starts from material depicting real abuse.

Report counts capture only part of the load, because someone still has to open, assess, and route what arrives. Internet Crimes Against Children (ICAC) task force officers in New Mexico testified in 2026 that they’re “drowning in tips” (Guardian, via Yahoo). Task force investigators told Bloomberg they sometimes crowd around one monitor trying to decide whether an image is real.

The cost to the people doing the review

AI-generated material is also more severe. The IWF rated 65% of AI-generated abuse videos in 2025 as Category A, its most severe tier, against 43% of non-AI videos (IWF).

That lands on investigators and forensic examiners. A study of U.S. personnel tied exposure to violent CSAM to higher post-traumatic stress symptoms (BMC Psychiatry). In a sample of 443 ICAC investigators, one in four reported high compassion fatigue, secondary traumatic stress, and burnout (Frontiers in Psychology).

The result is a double burden. Real victims wait longer to be identified, and the people working to protect them spend more hours exposed to severe material.

Reality Defender’s work with HSI

In 2025, Reality Defender began working with the Victim Identification Program at the Homeland Security Investigations (HSI) Cyber Crimes Center, under a Cooperative Research and Development Agreement. Reality Defender worked with a highly experienced team that needed to separate AI-generated material from real abuse in seized evidence and prioritize the cases involving real children. Efficiency of these investigations could not have higher stakes.

HSI’s casework shaped how the product works.

• Analyze the full frame. We run face-focused and full-frame models side by side, so the analysis covers the whole image, not only the face.
• Show the signals behind each result. Each result shows the overall score and conclusion alongside the individual model signals behind it.
• Multiple model results. Results with multiple reinforcing models increase the likelihood the reports provide strong, trusted signals for investigators and support a conviction.

Detection doesn’t remove the human from the loop. Instead it helps analysts increase operational efficiency and prioritize real material during a high stakes and time sensitive operation.

Increase investigation efficiency

The tools used to create AI-generated CSAM will continue to get cheaper, more widespread, and more extreme. Investigators cannot keep up with human eyes alone. Synthetic media detection helps them sort seized media early, so files that may depict a real child reach a human reviewer sooner.

Reality Defender supports law enforcement and child protection agencies around the world working on this problem. If your unit is facing it, talk to our public sector team.