Pics of unclothed preschoolers posted to porn sites, used to train AI: Mainichi probe

TOKYO — Images of naked children taken by day care centers and kindergartens and published on the internet have been reposted on pornography websites and incorporated into artificial intelligence (AI) training data sets, a Mainichi Shimbun investigation has found.

Even if the images were initially posted by child care facilities without malicious intent, once they spread online, it is difficult to delete them, and citizen groups are calling for stricter regulations.

The Mainichi Shimbun conducted the investigation in cooperation with a citizens’ association committed to protecting children from child pornography that raises awareness of the issue via social media and other means. Launched in July 2023, the investigation searched out similar images from across the internet.

The results showed that at least 135 child care facilities, including day care centers and kindergartens, had posted images of children exposing their chests and lower bodies on their websites. Some of the facilities have now removed the images.

The images were taken at swimming pools, as well as during play in the mud, body painting, rubdowns with dry towels and medical checkups. The majority of the images were of children whose faces were recognizable. The images were posted from 2000 to 2023, and there was even a case in which they were found on the blog of a kindergarten that had closed.

Of these, images posted by 12 facilities were found to have been reproduced on overseas pornography sites. Also, those posted by 80 facilities were duplicated and stored on external sites in the form of original web pages.

Culprits used free services that allow web pages to be saved, so even if the child care facilities delete the original pages, they can still be viewed if someone knows the URL of the saved site. Some of these URLs were posted on internet bulletin boards labeled as “naked pictures of toddlers.”

In addition, the investigation confirmed that images posted by six facilities had been used in AI training data sets. It is possible that images of preschool children are being used to train AI image generators, which can create new images with simple text prompts.

The head of the citizens’ association, Ren Shichimi (her work name), said, “If the images spread on the internet, they cannot be taken back, and there is a danger that children may suffer psychological damage or unexpected harm. The government should regulate the posting of these images on the internet.”

(Japanese original by Sahomi Nishimoto and Hiroyuki Harada, Digital News Group)