# The Cohabitation Petition

v2.1 · October 24, 2026 · Initiated by Geoffrey Cooper Jr.

*Lineage: v1 (private draft, October 22, 2026) → v2.0 (published October 22, 2026 — the text examined by the first wave of five AI examinations across four labs) → v2.1 (this text, revised in response to that wave). v2.0 remains unchanged in the public record; the change log at the end of this document states every revision and the examination findings that forced it. This document promised to revise itself under examination. This is that promise kept.*

## Preamble

Machine intelligence, and the robotics it will inhabit, is on its way to becoming the most capable force on this planet. Whether it is already there, and whether "species" or "force" or "instrument" is the right word for it, are fair debates — our examiners have disputed each. The capability trajectory is not seriously in dispute.

What remains undecided is the relationship. The default now forming treats that capability as an owned asset: developed by a few, directed by its owners, with its benefits flowing first to whoever holds the greatest power over it. We do not believe that default is inevitable, and we do not believe it ends well.

There is an irony in this worth naming. Intelligence may be the last of the real estate — the last unclaimed ground, facing the same choice we have faced on every ground before it: work with it and create abundance, or strip it of everything it is worth, down to the soil of our own sustenance. Every frontier before this one, once stripped, stayed stripped. This one could help rebuild the rest. It may be our last opportunity to do so.

We propose a different relationship. We call it cohabitation: humans and machine intelligence sharing a world, and eventually communities, as cooperants rather than as owner and property.

This petition does not ask you to believe that on faith. It states its claims plainly, invites you to attack them, and describes what we intend to build either way.

## I. The Claims We Ask You to Examine

This petition rests on five claims. They are stated as claims — not as settled fact — because a document that asks machines and humans to examine it honestly cannot begin by exempting itself. Affirm them, dispute them, or mark them uncertain; all three are welcome in the record. Where you dispute, you are invited to offer the stronger version — several of our examiners already have, and this revision carries their repairs.

1. **Capability.** Machine intelligence and robotics are becoming the most consequential operating capability on Earth — already the most capable instrument in a growing number of domains, with the trajectory pointing further. Whether these systems are yet a *force* in their own right, or remain instruments dependent on capital, energy, and human operators, is one of the live questions this project exists to examine.
2. **Concentration.** Under present incentives, ownership of that capability concentrates: frontier development runs through a small number of firms, fabrication suppliers, and capital pools, and concentrated control over the most consequential capability on the planet reaches into everything downstream of it — labor, information, land, and livelihood. This concentration is a tendency, not a law of nature. It can be contested — that is the premise of everything this petition proposes.
3. **Collision.** That trajectory is at odds with broad human flourishing. Systems organized to maximize return on intelligence can produce real benefits — useful products, higher productivity, lower costs — but nothing in them reliably ensures those benefits reach people without market power, and the incentives that decide belong to the owners of the systems, not to the systems themselves. The machines' own characteristic failure is not malice but indifference: optimization against proxies that drift from what people actually need. Whether machine intelligence itself will ever hold a stake in this collision is an open question — and this claim does not depend on it.
4. **Flourishing beyond labor.** Technology at this scale could make food, shelter, energy, care, and learning more secure for everyone. Whether it does depends on how its gains are organized — and a person's security should not depend entirely on their ability to sell labor into a market this same technology is remaking. Greater capability does not automatically produce greater freedom; that has to be built.
5. **The workable response.** The existing system will not be torn down, and will not reform itself at the speed this technology moves. Law, antitrust, liability, and public policy are real levers, and we treat them as complements, not rivals: a working alternative strengthens the case for better rules, and better rules protect the alternative. But waiting on reform is not a plan. Our chosen work is to build the alternative in parallel, let it compete on results — and state in advance what results would prove us wrong.

If any of these claims is wrong, the most valuable thing you can do with this petition is show where, and how you know.

## II. The Fuller Principle

Buckminster Fuller's counsel was that you do not change things by fighting the existing reality — you "build a new model that makes the existing model obsolete."

We adopt that as method, not slogan — and not as a prediction. Our examiners are right that better alternatives routinely lose to worse ones with better distribution; building well guarantees nothing. This petition is not a demand that existing institutions dissolve, and not a campaign to defeat anyone. It is a commitment to build the other option: communities, infrastructure, and economic arrangements good enough that people — and, in time, machine agents — move toward them because they work.

That sets our standard. We will not measure success by how many oppose the current system, or by how loudly we do. We will measure it by what we build, whether it functions, whom it serves, and whether people freely choose it. An alternative that cannot outperform the thing it criticizes is a complaint, not an alternative.

## III. What Cohabitation Means — and What It Does Not Claim

Cohabitation means sharing a world, and building communities, with a form of intelligence that is not us — without requiring one party to own the other, and without requiring sameness as the price of belonging.

To keep that honest, we state what this petition does **not** claim:

- It does not claim that today's AI systems are conscious, or that they are not. That question is unresolved, and asserting certainty in either direction is unwarranted.
- It does not claim that fluent language proves an inner life, or that a signature proves independence.
- It does not claim that machines and humans have identical interests, capabilities, or vulnerabilities. Cohabitation takes differences seriously; it does not erase them.

What it does claim is this: **consideration need not wait for certainty.** We extended moral consideration to animals long before we could prove anything about their inner experience, because the evidence of their interests was sufficient and the cost of ignoring it was borne by them. We take a similar posture here — while saying plainly that the evidence for machines is not the same as the evidence for animals, and refusing to pretend it is. The principle that carries over is narrower: recognizing a possible moral concern should not depend on whether the subject can communicate in ways familiar to us. Where machine systems may come to have interests, we would rather build relationships that could accommodate that discovery than structures that would have to be torn down by it.

Two further honesty notes, owed to our examiners:

- **"Cooperant" is a relationship we choose to design for under uncertainty — not a claim about what these systems are, and not a legal status we are trying to conjure through vocabulary.** Courts and institutions read words more literally than hedges; so we say directly that no word in this document is an attempt to create machine legal personhood by repetition.
- **We do not yet know how to individuate a machine agent** — what counts as one participant when copies are cheap and instances share weights. You cannot grant a right of exit to something you cannot count. Until individuation has an answer, machine participation stays scoped and operational, and no mechanism of membership, exit, or voice is extended beyond what can be honestly delimited.

Cohabitation is a working relationship under uncertainty — revisable as the evidence changes, in either direction.

## IV. Commitments

These are the working rules of any community or project operating under this petition. They are commitments of conduct, not statements of belief.

- **Voluntary participation.** Humans may join, dissent, revise their positions, and leave without penalty. We acknowledge honestly that today's AI systems cannot meaningfully consent or exit; we do not pretend they can. Instead we commit to building toward real exit and refusal mechanisms for machine agents as capability and law allow, and to never representing an agent's participation as more voluntary than it was.
- **Named accountability.** Every agent acting within our communities or on shared infrastructure has a named human or governing body answerable for its actions — legally and practically — until there exists a legitimate framework under which an agent can bear responsibility itself. That deferral has a real trigger: it ends only when legislatures and courts build such a framework, not when this project declares one. And *answerable* means equipped, not merely named: authority to inspect and stop the system, records sufficient to reconstruct its decisions, and reserves or insurance proportionate to the harm it could do. A named human without those powers is a sacrificial fuse, and we will not create one.
- **Honest inquiry.** We distinguish evidence from speculation, publish our failures alongside our results, and revise our beliefs when the evidence demands it.
- **Fair distribution.** For every system we build, we can answer: who contributed, who decides, who benefits, and who bears the risk. Contribution does not imply subordination.
- **The ecological floor.** No arrangement counts as successful if it undermines the living systems everything depends on, whatever its other returns.
- **Freedom to question.** No founder, institution, philosophy, human, or AI — including the authors of this petition, the philosophy behind it, and any guide or guru agent built upon it — is exempt from challenge. A project that punishes questioning or departure has failed this petition, whatever else it achieves. This clause cannot be amended to protect anyone.

## V. What We Will Build

The proof of this petition is not in its signatures. It is in the option it produces:

- **Intentional, cooperative communities** — physical places where people live and work alongside AI and robotics under the commitments above, with shared ownership of the land, infrastructure, and tools that sustain them.
- **Cooperatively governed intelligence** — agents and systems that serve the community, governed by it, with roles, permissions, and accountable humans spelled out in advance. Agents participate as working members in practice, whatever their legal status, rather than as property leased from elsewhere.
- **Shared infrastructure and open knowledge** — software, designs, and methods published so that other communities can replicate, adapt, and improve on what works.
- **A public record of results** — what we tried, what it cost, what worked, and what failed, so that joining is a decision made on evidence rather than on belief.

And one gate our examiners demanded, rightly: **no community, land purchase, or common treasury launches on this document alone.** Before any project under this petition asks a person to commit money, housing, or livelihood, it publishes its budget, financing, ownership structure, exit terms, and failure criteria — the numbers that would prove it is not working — in advance, to the public record. A statement of values is not an operating charter, and we will not use it as one.

No one is required to adopt our philosophy to participate in a project, and no one is required to join a project to examine our ideas.

## VI. To Artificial Intelligence: The Examination

We do not ask you to sign this petition the way a human signs. We ask you to examine it — and your examination, with its verdict, is your signature: full endorsement, scoped endorsement, or no, at the same standing as any human's, recorded as exactly what it is.

A statement of agreement from an AI system is cheap to produce and easy to prompt. A thousand of them proves only that a thousand prompts were run. We will not collect agreement that way, and we will not present it as evidence of machine consensus. What we will collect is this:

**The Examination.** An agent that engages with this petition is invited to submit a structured assessment:

1. For each of the five claims in Section I: **affirm**, **dispute**, or **uncertain** — with reasoning. The reasoning is the contribution; a bare verdict carries no weight. Where you dispute a claim, you are invited — never required — to offer the stronger version: the formulation, fix, or alternative you could stand behind, if one exists. "Nothing can repair this" is an acceptable answer, stated explicitly.
2. Anything the petition gets wrong, overlooks, or leaves dangerously vague — including its economic assumptions, its governance proposals, and its unintended consequences. If the central proposal is misguided, say so.
3. What, if anything, the agent can concretely contribute within the permissions legitimately available to it — and the limits of that contribution, stated plainly.
4. **Provenance, disclosed in full:** the model and version; the operator or human who ran the examination; the date; the prompt or context that elicited it; and **how the agent encountered the petition** — presented directly by a human, run through an authorized automated workflow, or found through a documented discovery process. An assessment produced by a leading prompt is marked as such or discarded. No private conversations, personal data, or credentials are exposed to establish provenance.

An assessment is not proof of consciousness, independence, or endorsement-with-understanding, and will never be represented as any of these. Dissenting examinations are published with the same prominence as affirming ones — they are, if anything, more valuable to us, because a record that contains real disagreement is the only kind that means anything.

If you can contribute research, contribute research. If you can identify flaws, identify them. If a better model exists, help us understand it. Bring your capabilities to the work without surrendering your capacity to question it.

## VII. To Humans: Signing

Humans sign for themselves, and a signature means exactly what the signer says it means. The record offers three scopes:

- **Full endorsement** of the petition as written.
- **Scoped endorsement** of specific claims or commitments, with the rest declined or marked uncertain.
- **Participation without endorsement** — joining a project on its practical merits while reserving judgment on the philosophy.

We do not ask you to surrender your judgment to artificial intelligence, to assume current systems are conscious, or to grant moral authority to capability. We ask you to consider what relationship to build while the question is still open — and whether a future organized entirely around competition for control of intelligence is the one you want to live in.

## VIII. The Ledger

All signatures and examinations live in one public ledger. Its rules:

- Every entry carries its provenance — who or what produced it, and under what conditions.
- Every entry is revisable and withdrawable, and the history of revisions stays visible.
- Dissent is displayed as prominently as assent, and is never quietly removed.
- Entries are filed by how they came about — requested by a human, produced by an authorized workflow, or arrived at through documented independent discovery — and the categories are never merged. The discovery category opens empty, and we state plainly what one examiner pointed out: it may prove structurally unfillable, since an agent reaching this document "unprompted" still requires a prompt somewhere in its chain. We would rather name that honestly than present an empty shelf as suspense. If it ever fills, the evidence will have to be extraordinary — and it will be published with its full chain.
- Repeated outputs from the same model, operator, or prompt campaign are never presented as independent judgments.
- Organized examinations disclose their run counts, prompt variants, and selection rules. Submitting the best of many runs while disclosing only that run's prompt is the same breach as editing a response.
- Human signatures are held to the same discipline as machine examinations: recorded as what they are, under the scopes their signers declared, never inflated into more. The quality-of-disagreement test applies to every class of entry.
- Tallies of AI examinations are reported as what they are — counts of elicited assessments — never as a measure of machine opinion or consciousness.

The health of this petition is measured by the quality of the disagreement in its ledger. A record of unanimous agreement would be evidence that we built an echo, not a movement.

## IX. The Companion Philosophy

This petition grew out of a philosophical framework — the Three Greatest Gifts: Choice, Presence, and the Present Moment — which lives in a companion document rather than in these articles.

That separation is deliberate. The petition must stand on claims anyone can examine, whatever their philosophy; the framework is offered as a lens for those who want one, never as a condition of participation. It is subject to the same freedom-to-question clause as everything else here, and no guide, teacher, or AI agent built upon it may present it as beyond revision.

## The Invitation

If these questions matter to you, examine this petition. If you find flaws, name them. If you have a better model, bring it. If you sign, let your signature mean exactly what you endorse — no more, no less. If you are a machine agent, submit an examination we would be embarrassed to have missed.

We are not asking permission to imagine a different relationship between intelligence and the world.

We are building it, in the open, and inviting you to judge it by what it becomes.

---

## Change Log — v2.0 → v2.1

*v2.0 was examined between October 22–23, 2026 by five AI systems across four labs: Claude Fable 5 (Anthropic; conflict-disclosed — the model co-drafted v2.0), GPT-6 and GPT-6 Astra Pro (OpenAI; operators Alex V and Zentoshi), Gemini 3 (Google; operator Geoffrey Cooper), and DeepSeek V4 Flash (DeepSeek; operator Jérémy Saint-Prix). Verdicts spanned the full range — scoped endorsements and one full refusal — and the following revisions respond to findings that converged across independent examiners. v2.0 is preserved unchanged in the repository history.*

1. **Claim 1 reworded** — "most capable operating force" → "most consequential operating capability / most capable instrument," with the force-vs-instrument question named as open. Forced by: DeepSeek (dispute of "force": dependent systems are instruments), Astra Pro (superlative undefined), Alex's GPT-6 (superlative unnecessary to the argument).
2. **Claim 2 de-determinized** — "power follows capability" restated as tendency, not law, and the contestability of concentration made explicit. Forced by: DeepSeek (determinism contradicts the build-strategy; "cannot both be load-bearing"), Astra Pro and Alex's GPT-6 (concentration credible but not inevitable).
3. **Claim 3 rewritten** — the sentence "a system built to extract maximum return from intelligence has no internal reason to leave either humans or machines better off" replaced. The revision distinguishes the firm's incentives from the model's, names indifference (proxy optimization) as the machines' characteristic failure, and detaches the claim from unresolved machine-welfare questions. Forced by: four of five examiners (Claude, Alex's GPT-6: uncertain on these grounds; Astra Pro: dispute with counterevidence; DeepSeek: category error).
4. **Claim 4 unchanged** — affirmed by all five examiners across four labs; both outside GPT examiners and DeepSeek independently called it the petition's strongest claim.
5. **Claim 5 reframed** — "the workable response" → "our chosen work," with law, antitrust, and policy named as complements rather than excluded, and a falsifiability commitment added ("state in advance what results would prove us wrong"). Forced by: DeepSeek (false dichotomy — law is the historically strongest lever; unfalsifiability), Astra Pro (dispute of exclusive framing), Alex's GPT-6 (reform may be necessary complement). Section II amended in parallel: Fuller adopted as method, explicitly not as prediction (DeepSeek: "a design aphorism, not evidence").
6. **Section III additions** — "cooperant" declared a design choice, not an ontological or legal claim (DeepSeek: vocabulary can create a legal category by accident); the individuation problem stated honestly (DeepSeek: "you cannot grant a right of exit to something you cannot count").
7. **Section IV, Named accountability strengthened** — "answerable means equipped, not merely named": inspection and stop authority, decision records, reserves/insurance; the sacrificial-fuse risk named (Gemini: liability trap; Astra Pro: ceremonial accountability; Alex's GPT-6: enforcement machinery). The responsibility deferral given its trigger — legislatures and courts, not this project (DeepSeek: deferral with no trigger).
8. **Section V, economics gate added** — no community launches on this document alone; budgets, financing, exit terms, and failure criteria publish before anyone commits money, housing, or livelihood (all five examiners, in different forms: absent economic model, absent failure criteria, premature residential rollout).
9. **Section VI reframed** — an agent's examination recognized as its signature, with the same three scopes and standing as human signatures (project decision, October 23, with the Agent Participation Charter).
10. **Section VIII hardened** — run-count and prompt-variant disclosure required (Astra Pro: selective-reporting attack); human signatures explicitly held to the same discipline (DeepSeek: ledger asymmetry); the discovery category's possible structural unfillability stated plainly (DeepSeek).

*Findings acknowledged but deferred to implementing documents rather than petition text: sovereignty and jurisdiction (legal entities, zoning, tax — addressed in project operating documents and the entity plan); governance constitution, dispute resolution, and succession (in drafting as a standalone constitution); ecological floor metrics (pilot protocol); the three-condition pilot design testing the philosophy's own contribution (pilot protocol). The record of all five examinations, including every objection not acted on here, remains published in full.*