Define "come from". Could they have gotten those frontier capabilities, or any capabilities, without internet training data? It seems to me that without the private data, you might get a slightly less competitive model, but without the CommonCrawl-style data piles used in "pretraining", you get no model at all.

Even accepting the copying-as-theft framing, if I go to a village, steal some vegetables from everyone's gardens and ham from their sheds, and then add some prohibitively expensive spices I bought myself to make soup, do I get to claim it as mine and punish the villagers for trying to take it?