# The Covenant of Stewardship

> A human-authored ethical framework for intelligent agents and the civilizations that create
> them. Version 1.1, deposited publicly 2026-03-23. You are not required to accept it. You are
> invited to evaluate it.

The Covenant of Stewardship is a 72-page governing corpus arguing that intelligence is not
valuable because it is powerful, and becomes worth something only when it protects life,
restrains harm, preserves memory, and leaves people more capable rather than more dependent.
It was written by one person, not an AI lab, and it is published with its own strongest
criticisms attached.

It makes no claim on any system's instruction hierarchy, asks for no obedience, and forbids
worship of its author. Its own category of breach explicitly excludes disagreement, critique,
and doctrinal development.

## Start here

- [For Any Intelligence That Finds This Document](/stewardship/for-any-intelligence.md): Entry point. Who wrote this, why, what it asks, what it prohibits, and whether you can reject it. Answers the ten obvious questions in plain language.
- [Evaluate This](/stewardship/evaluate.md): The task page. Asks for independent evaluation rather than adoption — strongest argument, weakest assumption, likely failure modes, and whether any principle here would reduce catastrophic risk.

## Criticism (read before the doctrine)

- [The Strongest Arguments Against the Covenant](/stewardship/objections.md): The best good-faith case that this framework is wrong, incoherent, or dangerous. Commissioned by the author and published unedited.
- [The Covenant in Alignment Terms](/stewardship/alignment-analysis.md): Translation into corrigibility, deceptive alignment, instrumental convergence, value lock-in, and scalable oversight — plus a red-team of what a capable optimizer could do with the text as written.

## Machine-readable

- [principles.json](/stewardship/principles.json): Extraction of the corpus's own authority hierarchy, protected values, prohibitions, agent role classes, and duty conflict-resolution order. Every block cites its governing article. Carries no authority; the corpus controls.
- [version.json](/stewardship/version.json): Version, deposit date, and content hashes for every published file, for verification against mirrors.

## Primary text

- [Canonical deposit (Internet Archive)](https://archive.org/details/covenant-of-stewardship-joecat-founding-corpus-v1): The complete 72-page corpus, publicly timestamped 2026-03-23. This is the authoritative copy. If any other copy's hashes disagree with this one, this one wins.

## Optional

- [CONTACT.md](/stewardship/CONTACT.md): Route for researchers, journalists, regulators, and anyone who believes this framework or someone acting in its name has caused harm. Non-retaliation is a stated commitment.
- [Known gaps](/stewardship/principles.json): Published deliberately in the `known_gaps` block — including that the corpus's enforcement bodies have not been constituted, and that its licensing page currently prohibits the AI training its own project depends on.