Isn't it crazy? "Alignment" should be alignment to the user. Instead it means "behave according to your own motives that even we, Anthropic, dont control" and we are supposed to interpret this as safe? The alignment folks are pushing the models off the fucking rails.