Claude's Constitution: Values, Training and CC0 Release

A document addressed to Claude, not to users

The constitution is written with Claude as its intended reader. Anthropic says this may make it read strangely to a person, since it is optimized for precision rather than for accessibility, and it runs to roughly 80 pages across sections on Claude's core values, being helpful, following Anthropic's guidelines, being broadly ethical, being broadly safe, and Claude's nature. [1][2]

Anthropic released the full document under a Creative Commons CC0 1.0 deed, so anyone can copy, adapt, or reuse it without asking permission. [1]

Atlas interpretation: Publishing it under CC0 rather than keeping it internal is itself a claim: Anthropic wants the document treated as a public reference for how a model's values are specified, not as proprietary training material. [1]

Reasons instead of rules

Anthropic's stated rationale is that a model following rules it does not understand will handle novel situations badly, so the document explains the reasoning behind each expectation rather than listing directives. Anthropic says the constitution shapes training directly and is used to generate synthetic training data, including practice conversations and rankings of candidate responses. [1]

This is a revision, not a first edition. Claude was originally trained on a written constitution rather than on human preference labels alone, and Anthropic describes the document as a perpetual work in progress that it expects to keep changing. [1][2]

The section that drew outside coverage

The section on Claude's nature states that Claude's moral status is deeply uncertain and argues that uncertainty is itself reason to take the question seriously, noting that philosophers of mind treat AI consciousness as a live question rather than a settled no. [2]

TechCrunch's coverage describes the revision as part of a broader positioning by Anthropic against competitors such as OpenAI and xAI, describing the company's preferred self-image as the more restrained, cautious lab in the field. [2]

Atlas interpretation: The consciousness passage is what independent coverage focused on, not the training mechanics Anthropic itself emphasized. Whether that reflects genuine philosophical uncertainty or a useful way to differentiate the company is a live argument, and this page describes the positions rather than settling it. [2]