AI startup Anthropic has revealed the written “constitutional” principles used to train its systems. These include principles taken from the UN Universal Declaration of Human Rights and Apple’s terms of service.