In an analysis of approximately 3,600 messages from AI power user group chats, 22% of messages identified trust as their primary concern regarding AI agents