It's an image every few seconds. Not that piracy is currently even interested in tech that reencodes the content.
And for training, copyrighted stuff is already everywhere; AI tools seem to be limited on the output side rather than raw training data.