If we see court judgements start to go copyright owners way, we will also see a scramble from AI companies to buy the few publishers with enough data to be worth buying, and to create works for hire to replace the rest.
In the long run a copyright ruling like that will be a boon for OpenAI and all other players with deep enough pockets to do so, and massively harm everyone else who will suddenly find it far harder to build models legally.
So that would mean articles from the 1920s, provided that the authors of those articles have been dead for 70 years, or longer in some other countries.
