OpenAI and Google trained their AI models on text transcribed from YouTube videos, potentially violating creators’ copyrights, according to .