Yep and many (as I do) claim open-source will help mitigate this problem, but these models actually need to be fully transparent about the datasets (and licenses) used for training. Plagiarism shouldn’t be normalised.