How Image Generators Are Trained on Non-Consensual Photos

In this episode, Lucas and Luna examine a troubling aspect of generative AI: the use of non-consensual intimate images in training datasets. They focus on a 2025 Stanford Internet Observatory report that found more than 3,000 such images in the LAION-5B dataset, which has been used to train models like Stable Diffusion. The hosts discuss how this happened, why it matters for privacy and consent, and what researchers are doing to scrub datasets and build better auditing tools. They also explore the legal landscape, including the UK's Online Safety Act and proposed US legislation, and ask whether the AI industry can self-regulate or needs stronger enforcement. A thoughtful conversation about the hidden costs of open data and the ethics of scraping the web at scale. #GenerativeAI #NonConsensualImages #DataEthics #StanfordInternetObservatory #LAIONDataset #PrivacyViolation #AITrainingData #ConsentInAI #StableDiffusion #OnlineSafetyAct #WebScraping #AIEnforcement #Technology #AIEthics #FexingoBusiness #BusinessPodcast #DeepfakeLegislation #ImageGeneratorBias Keep every episode free: buymeacoffee.com/fexingo