“Anthropic's constitution for its AI model Claude includes a direct apology to the AI in the event that it is a 'moral patient' experiencing unnecessary costs from its training or operation.”