They don't need to. They have implicit, automatic copyright on the published original works. The lawsuit shows examples of GPT using the NYT work verbatim.
I don't think they should have a right to the records of people who lived before simply bc they underpaid, over worked human journalists to exploit their creativity - largely to make the "Newspaper Barons" rich.
I don't think they ought to have implicit, automatic copyright on older stuff. How old is my only debate.
What good is old news really? Real old is good for college papers but news from 4 years ago... who cares?
OpenAI does not distribute it models, so the violation of copywrite if found to be a violation will be limited to the amount of people who queried the LLM in the same way that NYT did.
How that copyright shakes out in the end (what your statements focus on) is yet to be determined.
I'm saying that between "you're violating copyright law" and "you're violating copyright law and you supposedly agreed you a license that contractually prevented you from doing this", the second sounds like a stronger claim.
So existing copyright law was sufficient for written media, but didn't quite work for software. OTOH it seems that GPL-style copyleft licenses are being used more by writers and artists, so it's not like these issues exist in a vacuum. I expect copyleft will be more appealing to artists going forward given Midjourney et al.