In 2025, Amazon.com discovered hundreds of thousands of images and files it believed depicted child sexual abuse while reviewing data intended for training its artificial intelligence models. The company removed the content before training the models and reported it to the U.S. based National Centre for Missing and Exploited Children (NCMEC). However, Amazon has not shared detailed information about the data’s origin, making it harder for child protection officials and law enforcement to trace the material’s origin.
NCMEC, an organisation created to collect and forward tips about child abuse to law enforcement, has seen a dramatic rise in reports related to AI. According to the centre’s tracking, the number of AI-related reports submitted in 2025 was at least fifteen times higher than the previous year, with the vast majority coming from Amazon.
Technology companies competing in the race to build powerful AI systems often rely on large-scale datasets derived from publicly available online content. These datasets can inadvertently include illegal or harmful material if not properly vetted. Experts warn that training AI on such content not only poses serious ethical and legal risks but also risks reinforcing harmful patterns of output, including the inadvertent generation or reproduction of similar material, if the training process is not carefully controlled.
@ANP | BLOOMBERG NEWS | NEWS BRAINPORT
