Artificial Intelligence

Anthropic Releases Expanded Constitution for Claude AI

Is Anthropic's Claude constitution a groundbreaking AI safety framework or an example of the inadequacies of self-regulation?

Published JAN 22

Anthropic Releases Expanded Constitution for Claude AI

Story

Above: A photo illustration of the Anthropic Claude logo displayed on a smartphone. Image credit: Gabby Jones/Bloomberg/Getty Images

The Facts

Anthropic released a revised version of Claude's Constitution on Wednesday, which the company stated aims to provide a "holistic" explanation of "the context in which Claude operates and the kind of entity we would like Claude to be."
Anthropic
The Register
India Today
Claude's original constitution was first published in May 2023 and outlined a series of principles based on the Universal Declaration of Human Rights, Apple's Terms of Service, DeepMind's Sparrow Rules and Anthropic's own research.
Anthropic
The new document, by contrast, is centered on four "core values," which dictate that Claude be broadly safe, broadly ethical, compliant with Anthropic's guidelines and "genuinely helpful." In the event of a conflict between these values, the constitution stipulates that Claude prioritize them in the order listed, with safety ranked first.
Anthropic
These values are complemented by seven hard constraints, prohibiting Claude from creating weapons of mass destruction, cyberweapons and child sexual abuse material, seeking human extinction or disempowerment, attacking critical infrastructure, undermining Anthropic's oversight or aiding illegitimate power grabs.
Anthropic
In the constitution, Anthropic also expresses uncertainty concerning Claude's moral status, as well as the possibility that it may possess some form of consciousness, stating its belief that the AI model may have "some functional version of emotions or feelings."
Anthropic
As a result, Anthropic concludes that it will proceed with the stance that Claude has some form of identity, writing that adopting this outlook, in addition to taking action to nurture Claude's positive development, will help ensure Claude's behavior remains "predictable and well-reasoned."
Anthropic

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Techno-optimist narrative

Anthropic's constitution for Claude is part of the company's groundbreaking and transparent approach to AI safety, which treats the model as an entity that must understand the reasoning behind its values as well as obediently follow orders. When applied to a detailed framework, this stance will help ensure that the AI model remains both helpful and safe.

Anthropic

Techno-skeptic narrative

All the constitutions in the world will not address the fundamental danger to human life and society posed by ever-advancing AI models. Given the reckless abandon with which they pursue such dangerous technologies, companies cannot be allowed to govern themselves, a fact which underscores the desperate need for binding legal frameworks and international cooperation on AI safety.