I think the point is that they're akin to a watermark.
Even before the current AI boom, plenty of artists have wanted to showcase their work/prove that it exists without necessarily making the highest quality original file public.
For example in accounts on image sites that are exposed to suspected scrapers but not to others. Scrapers will still see the real data, but they'll also run into stuff designed to mix up the training process.
It's a bad tradeoff.