Should AI Assistants Show The Stop Rule Before First Run

The first-run screen is where this either feels safe or sketchy. Before an assistant gets files, apps, or deployment access, show the user three plain buckets: can do alone, can draft only, must ask first. Then add the one sentence people actually need when tired: what will make it stop. Not a permissions wall. More like a boarding pass for the run: reading these folders, drafting in these apps, never sending or deploying without approval, stopping if the source is missing or the cost changes. A new user should not have to learn the product’s safety model by hovering over tiny lock icons after something scary already happened.