Amazon points out that the number of child sexual abuse contents in AI training data reported is 'unusually large' and the source is unknown
Amazon is building a system to detect child sexual abuse material (CSAM) from datasets used to train its AI, and when it detects it, it reports it to the National Center for Missing and Exploited Children (NCMEC). However, it has been reported that NCMEC is struggling because Amazon does not know the source of this CSAM.
Amazon Found Child Sex Abuse In AI Training Data - Bloomberg
Amazon is committed to preventing CSAM across all of its services, including AI learning, and is actively eliminating it using machine learning, keyword filters, automated detection tools, and human review. In 2024 alone, the company submitted a total of 64,195 reports to the NCMEC.
Amazon's efforts to combat child sexual abuse material
https://www.aboutamazon.com/news/policy-news-views/amazon-csam-transparency-report-2024
In a 2026 interview with Bloomberg, the NCMEC reported that 'more than 1 million AI-related CSAM cases were reported in 2025, the majority of which came from Amazon.' They also noted that 'only a few cases came from other companies, showing a significant disparity,' and that 'Amazon does not share details about the origins of its data, which could hinder law enforcement's efforts to identify perpetrators and protect victims.'
According to Fallon McNulty, executive director of CyberTipline, an NCMEC hotline, while Amazon is not required to report the origin of data, other companies share more detailed information. NCMEC has told Amazon that 'such reports are not going to do anything,' because it can't do anything without knowing the source. McNulty said, 'Simply reporting something you've found without providing any actionable information is not helpful to the child protection field as a whole.'
An Amazon spokesperson said the training data comes from external sources and that Amazon does not have detailed information that would be useful to law enforcement in their investigations, adding that 'as of January 2026, we are not aware of any instances where our models have generated CSAM.'
The number of AI-related CSAM reports to NCMEC exceeded 1 million in 2025, compared with 67,000 in 2024 and just 4,700 in 2023, showing an unusually high annual increase. While the fact that the majority of these reports came from Amazon could be interpreted as an indication that Amazon's detection system is superior to other companies, NCMEC believes that 'such a large number of reports received throughout the year raises many questions about the source of the data and the safeguards in place.'
According to Amazon, they set a low threshold to avoid accidentally missing CSAM, which has resulted in a high number of reports to NCMEC.
AI-related reports only make up a small portion of the total. The total number of CSAM reports, including those sent via private messages and posted on social media, reaches 20 million. Meta-related services, such as Facebook, Instagram, and WhatsApp, were the most common locations for detection. However, not all of these reports are necessarily confirmed as containing CSAM.
Related Posts:
in AI, Posted by log1p_kr