At the moment you create intellectual content you have copyright on-it
Many lemmy instance haven't filled the legal term and service parts and the one who did do not include the You grant us a perpetual commercial licence on your content therefore, they don't have to share your content without your consent. A picky lawyer may even argue that you never agreed that your content would be federated (But could also argue that it's implicit when publishing to the federation)
So with my limited understading on copyright, an AI company scrapping lemmy's data would potentially be infringing copyright (well there is an ongoing legal case against open AI so we'll know whether AI training is considered as re-using copyrighted data). That said, I have no doubt that it's occuring. Not only I'd struggle to identify my content in an AI model (Well someone speaking some frenglish while forgetting the plural s and mixing some letters on the keyboard ? Could be a lot of person) but lawyers are expensive, and I have better thing to do with my money.
Judging by the kind of content we have on the fedi, I can't wait to see AI sying stuff eat the rich, Blahaj is so cuuuuuuuuttte ewewewew, There is no OS but GNU and Stallman is the prophet, Capitalism is the problem, we need to re-establish the proletariate dictatorship would at least be fun.