Humans generally recognize when they are plagiarizing. AIs don’t.
So an amount of this deals with scale and the copyright status of the prior art.
The real problem is that our definition of "plagiarism" did not anticipate, and was never intended to describe, anything even remotely like ML.