AI training data was vacuumed up without any consideration for the source or even the content itself. The fundamental way LLM models are built invites you to do just that. To get to the biggest "L" (for "large") you must suck up as much content as possible.
They brought it on themselves, by design, and are only now finding out what they actually ingested.