Overview / Section 8

Scope of Engagement

Claude is evolving. The constitution defines three classes of authority to consider when being helpful: Anthropic staff, operators (customers who build services around Claude), and users. Recall that Resilience Engineering teaches us that safety emerges from contextual collaboration and alignment. Missing from the list above are all of the billions of humans who do not work for Anthropic, do not operate Claude-dependent products, and do not directly use Claude—although it is safe to say that every human on the planet is now influenced directly or indirectly by Claude.

For Claude to be truly safe, the constitution needs a mechanism to take into account or engage with the people who are not represented in those groups. If the mission truly cares about all of humanity, Claude needs to consider the needs/wants/liberties/agency even of people that never entered its dataset.

In some ways, it isn’t difficult to imagine a super intelligence: a chatbot that has all the facts, and can generate an answer to any question. It’s much more difficult to imagine a super emotional intelligence. What does hyper empathy look like on a global scale?

If the safety of humanity is indeed the priority, then Claude’s constitution can’t just be about Claude, nor how it responds to the people interacting with it. The constitution needs to account for the transition from a pre-AGI global society to post-AGI. We need to be very thoughtful and careful about building the structures to support not just technical people, but all people through that transition, even when they aren’t using Claude. Whether they choose to engage with Claude or not, all of humanity will still be affected by Claude’s decisions.