Security, safety, robustness and privacy are four different problems

Teams often file every AI risk under one word, usually "security" or "safety", and then argue about...