We all know diffusion models and LLMs were built by scraping the internet. This isn’t anything new. But what exactly is the harm in this specific scenario?