They're not conscious in the same way humans and animals are.

  Critically, nobody trained Claude to do anything like this; the behavior emerged on its own.

They clearly are trained to do so even if Anthropic didn't deliberately optimize for it. These models were trained on human text and knowledge. Of course they will talk about consciousness. If cats trained the models, all they'll talk about is sleeping, catching mice, and not on consciousness.