Adobe says the system is only trained on content that’s licensed or out of copyright — not work from artists across the internet.