In a recent breakthrough, artificial intelligence researchers announced that they have successfully removed over 2,000 web links containing suspected child sexual abuse imagery from a prominent database used to train popular AI image-generator tools.
The LAION research database, a significant resource for top AI image-making tools like Stable Diffusion and Midjourney, came under scrutiny after a report by the Stanford Internet Observatory exposed links to sexually explicit images of children within the dataset. This discovery highlighted the role these images played in the creation of realistic deepfakes involving children.
Following the report, the nonprofit Large-scale Artificial Intelligence Open Network (LAION) promptly removed the problematic dataset. Months later, after collaborating with watchdog groups and anti-abuse organizations, LAION released a cleaned-up version of the database for future AI research.
While LAION’s efforts were praised for significant improvements, Stanford researcher David Thiel emphasized the need to remove the “tainted models” still capable of generating child abuse imagery. One such model, an older version of Stable Diffusion, was taken down from the AI model repository Hugging Face by the company Runway ML.
The renewed focus on cleaning up AI databases coincides with global concerns about the misuse of technology in creating or disseminating illegal images of children. In San Francisco, a lawsuit was filed to shut down websites facilitating the creation of AI-generated nudes of women and girls, highlighting the urgency of addressing such issues.
Furthermore, recent developments involving the messaging app Telegram have prompted authorities in France to file charges against the platform’s founder and CEO, Pavel Durov, over the alleged distribution of child sexual abuse images.
