Most LLM chatbots don't push back when they should. When combined with situations like these, at a large scale, even 5 percent is abysmal, let alone 55 percent.