AI Training Data Contaminated with Child Sexual Abuse Imagery, Study Reveals
Stanford's Internet Observatory discovered that the popular AI image training dataset, LAION-5B, used by Stability AI and Google's Imagen image generators, contained links to child sexual abuse imagery. The dataset included at least 1,679 illegal images scraped from social media and adult websites. While the dataset does not store the images itself, it provides links and alt text. LAION, the nonprofit managing the dataset, temporarily removed it, emphasizing a "zero-tolerance" policy for harmful content. Stanford researchers recommended deprecating and ceasing distribution of models trained on LAION-5B, and US attorneys general have called for an investigation into the impact of AI on child exploitation and the prohibition of AI-generated child sexual abuse material.