All text written by a human in the US is automatically copyright the author. So if an engine trained on works under copyright is a derivative work, GPT3 and friends have serious problems.