You don’t need every post, just a collection big enough to train an AI on. I imagine it’s a lot easier to get data from the Internet Archive (whose entire mission is historical preservation) than from Reddit.

The thing I’m not sure about is licensing, but it seems like that’d the case for the whole AI industry at the moment.