This vague situation with copyright plays against open-source AI models who have to disclose the sources of training data, while closed-source companies can freely use pirated material and get advantage over open-source models.