> copy pasting it's training data

This is a total misrepresentation of how any modern LLM works, and your argument largely hinges upon this definition.