The issue is in federal courts and has not reached the supreme court. Most of the judges up to now have sided with Ai companies. They can read and digest any data apparently, the entire knowledge of humanity including whatever you post here and have ever posted your entire life. They can't download the stuff and save it though, so in effect they have to pay some licensing fee or constantly read and never store the data... don't take my word for it, check it out.
That's not what's being decided, it's the input of copyright content and so far it is leaning strongly towards the AI companies being able to do that and only getting in trouble for pirating the material first.
The rationale is part fair use and (more) part the process is transformative.
If the Ai can output enough to violate copyright, that's when an author can make claims. Until it does, it's just speculation that it might one day do that
If the Ai can output enough to violate copyright, that's when an author can make claims. Until it does, it's just speculation that it might one day do that
the NYT case is another, but that was an older model more prone to that, and text is easier to (eg.) hash and prevent with a post-gen guardrail