Does this technology also poison the well when there's already existing training data? If an artist's well established style that they've honed for years has al

Does this technology also poison the well when there's already existing training data? If an artist's well established style that they've honed for years has already been scraped, this isn't particularly useful unless this would also begin to disrupt the whole model of that artist's style in an existing training set. Better late than never, but if it can poison existing training data, that might be worthwhile.