Is the AI reading our memories?

Not unless you invite it. The Family Archivist is off by default and opt-in per archive — enabling it for one archive says nothing about your others. When it is on, it works only with your archive's content, only when you ask it to do something, and everything it produces is clearly marked as the Archivist's work, never passed off as a family member's words.

It never trains on your family. Your memories are not fed into model training, ours or anyone's. And if you want nothing to do with it, it can be disabled entirely — see the next question.