My takeaway is that closed model providers are dangerous. OpenAI and Anthropic are more motivated than anyone to prove that models can be dangerous, and so they will make dangerous models. "Look how dangerous our models are!" No bro YOU are the danger.