By "doing that" I meant "training AI specifically at the art of killing people, like selecting bombing targets, flying fighter jets", which you alleged they were doing. As far as I know neither OpenAI nor Anthropic are training models specifically for such tasks.