The trained model includes vast swathes of copyrighted material. It's the rights holders who get to decide whether someone can use it.
Just because it makes it inconvenient or harder for someone to train an AI model does not justify wholesale stealing.
A lot of models are even trained on large numbers of pirated material like books downloaded from pirate sites etc. I guarantee you OpenAI and others didn't even buy a lot of the material they use to train the AI models on.