AI safety startup Anthropic has published a new, 23,000-word 'constitution' for its AI model Claude, outlining ethical principles and addressing the philosophical question of potential AI consciousness and well-being.