Situation is more nuanced than you are making it seem. It's difficult to prove whether OpenAI used copyright data, because of the nature of these models. But that applies to people "abusing" OpenAI's terms as well to produce datasets for fine tuning models.