Amazon Discovers Massive Trove of Child Abuse Material in AI Training Data

The Discovery

Amazon has made a troubling discovery in its rush to develop artificial intelligence: the company reported hundreds of thousands of suspected child sexual abuse images found in data it collected to train AI models last year. This accounted for the overwhelming majority of more than one million AI-related reports received by the National Center for Missing and Exploited Children in 2025.

A Fifteen-fold Surge

The findings reveal a fifteen-fold surge in AI-related reports to NCMEC compared to the previous year. In 2024, Amazon and all its subsidiaries made roughly sixty-four thousand total reports to NCMEC. That number skyrocketed in 2025 as the company ramped up its AI development efforts. For comparison, NCMEC received just four thousand seven hundred AI-related reports in 2023 and sixty-seven thousand in 2024.

Reports Deemed Inactionable

While Amazon says it removed the content before using it to train models, child safety officials have criticised the reports as largely useless for protecting children. Fallon McNulty, executive director of NCMEC's CyberTipline, called Amazon's situation an outlier, noting the company provided very little to almost no information about where the illegal content originally came from, who shared it, or whether it remains available online.

Industry Comparison

NCMEC pointed to glaring differences between Amazon and other tech companies that scan AI training data. Google, Meta, OpenAI, and Anthropic said they also scan training data, but their AI-related reports came in much smaller volumes and included key details that allowed authorities to take action. Amazon uses what it calls an over-inclusive threshold for scanning, which yields a high percentage of false positives.

Calls for Transparency

David Thiel, former chief technologist at the Stanford Internet Observatory, called for greater openness from the industry regarding how companies gather and analyse data to train their models. Amazon maintains it is not aware of any instances of its models generating child sexual abuse material.