Probably the main risk would be people trusting what they think to be an automated system and trusting it to act for a specific purpose. That or people saying things they think should be private. I’m not saying it’s actually safe to interact this way with bots, but the trust expectations are different enough.