The trick here is to focus on imaginary safety from intentional AIs while ignoring the risks posed by real people using AI against other people.