The dataset might contain only (or be filtered for) confirmed email addresses, ie. where someone actually clicked the confirmation link. That would make this very unlikely.

Movements to abolish or reduce the scope of copyright existed long before the current “AI” hype. Keywords around it are: free culture, the Pirate Bay, “copying is not theft”, etc.

I’ve long been sympathetic to such and I don’t think “AI” changes anything about that.