The whole AI industry runs on the principle of "you can never prove that this output stole from this specific input, so what we are doing is totes legal". It's theft that's being legalized through fancy math - which is why I reject AI models that are trained on copyrighted data without the consent of the creator as immoral outright and do not want anything to do with people who use those models.
That doesn't make it ethical to do so, which is why we have AI disclosure - so that I, as a consumer, know that you are willing to steal from others as long as you know that you can get away with it, and can then decide that I do not want anything to do with you.