I am interested to see how copyright will stop companies from using copyrighted context pretraining AI model, then generating synthetic data by refining input d

I am interested to see how copyright will stop companies from using copyrighted context pretraining AI model, then generating synthetic data by refining input data thus making a much stronger and not so verbatim data. It's illegal to use copyrighted material this way, but detection seems difficult. I really doubt they could prove that such model was created using copyright work, unless someone leaks this information.