Spam and illegal content get in, but they can't get out.

Input from unknown sources gets stalled in a moderation queue and not served back out (except to authorized users of the relay, they can see it).  So if someone replies to you, you immediately see it, but others watching the conversation don't see it until it passes moderation.

This means I can moderate "lazily", I don't have to be on edge ready to quickly take something down.