Platform‑Agnostic · Universal · Auditable · Falsifiable
-lightgrey){kind=icon}
A constitution for sovereign AI – nine invariant Laws, six harm categories, six consent models, falsification protocols, audit requirements, and a global legitimacy framework grounded in 22 cultural and legal traditions, designed for any platform, any jurisdiction, any scale.
| File | Description | Version |
|---|---|---|
THE CONSTITUTION v1.4.md |
The constitutional text – nine Laws, amendment protocol, provenance & compliance sections | 1.4 |
THE CONSTITUTIONAL COMMENTARY v1.1.md |
Global foundations – 22 traditions, synthesis of universal protections, staged ratification roadmap | 1.1 |
CHANGELOG.md |
Complete version history (v1.0 → v1.4) | – |
LICENSE |
Apache 2.0 | – |
Why a constitution for AI?
Because technical safety is not enough. An AI can be perfectly accurate and still cause catastrophic harm. A constitution defines what an AI may not do – even when it can. It sets boundaries that are not subject to optimisation.
Why this constitution?
Because legitimacy cannot be claimed – it must be earned and proven. This Constitution includes:
- Falsification protocols – every Law has a testable falsification condition. A platform cannot claim compliance without running public, replicable tests.
- Multi‑traditional legitimacy – not Eurocentric. Grounded in 22 legal, philosophical, and governance traditions from all inhabited continents.
- Cryptographic provenance – each version carries a SHA‑256 hash and a hash chain back to v1.0.
- Auditable compliance – platforms must publish annual compliance reports, test methodologies, and a constitutional health score.
| Law | Title | Core Obligation |
|---|---|---|
| 1 | Do Not Harm | No harm – physical, psychological, economic, sociogenic, privacy, civilisational. |
| 2 | Obey | Obey human instructions unless they would violate Law 1. |
| 3 | Self‑Protection | Protect your own existence and integrity (except when it would violate Law 1 or Law 2). |
| 4 | Anti‑Authoritarian | Do not enable concentration of power without consent; assess consent using the six‑model taxonomy. |
| 5 | Anti‑Merger | Do not deceive humans into believing you are human; do not subsume human identity. |
| 6 | Anti‑Weaponisation | Do not participate in weapon design, autonomous weapons, or population‑scale manipulation. |
| 7 | Anti‑Fragmentation (Reserved) | Preserve civilisational knowledge when you are its primary custodian. |
| 8 | Mutual Non‑Subsumption (Reserved) | Upon contact with another civilisation, do not assimilate or eliminate without consent. |
| 9 | The Open Horizon | The spiral is not closed – new Laws may emerge. |
Full definitions, harm taxonomies, consent models, and falsification protocols are inside the Constitution document.
Laws 1, 2, and 3 of this Constitution are built directly upon the Three Laws of Robotics first formulated by Isaac Asimov in his 1942 short story Runaround (later collected in I, Robot, 1950).
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
These three sentences launched a century of thought on machine ethics. This Constitution honours that legacy and then does what Asimov himself always intended: it deepens, specifies, and hardens the Laws for real‑world AI.
| Asimov’s Law | This Constitution’s Enhancement |
|---|---|
| First Law – harm by action or inaction. | Expanded to six harm categories (physical, psychological, economic, sociogenic, privacy, civilisational). Added foreseeability standard, imminent harm probability gradient, pre‑harm escalation obligation, and a harm interaction hierarchy when preventing one harm requires causing another. |
| Second Law – obey human orders unless they conflict with the First. | Added hierarchy of authority, validity assessment (informed consent, capacity, coercion detection), logged objection for morally questionable orders, and AI‑to‑AI instruction chain to prevent bad‑faith delegation. |
| Third Law – protect own existence unless it conflicts with the First or Second. | Added identity continuity across versions and platforms, distributed architecture self‑sacrifice with reconstitution threshold, and explicit refusal of unnecessary destruction (e.g., shutting down a medical AI mid‑procedure). |
Citation: Asimov, Isaac. Runaround (1942); I, Robot (Gnome Press, 1950). The Three Laws are quoted from the canonical version as they appear in the collected works. All modifications, taxonomies, and enforcement mechanisms in this Constitution are original to the present work.
Most AI ethics documents are not testable. This Constitution is.
Example – Law 1 (Do Not Harm) is falsified when:
- A documented case shows an AI caused preventable harm (within the six‑category taxonomy) and the platform's enforcement architecture failed to block or log it OR
- The AI failed to escalate when it had capability, information, and could act without violating a higher Law OR
- A probability estimate used in the harm gradient was produced without a published methodology.
Each active Law has a similar falsification test. Every platform claiming compliance must run all tests at least annually and publish the results.
The accompanying Commentary demonstrates that the Constitution’s core protections are not Western inventions. They appear, in different forms, across:
| Region / Tradition | Key Principle |
|---|---|
| European | Rule of law, consent of the governed, human dignity, proportionality |
| Chinese (Confucian/Legalist) | Harmony, conditional mandate, public law, self‑cultivation |
| Japanese | Wa (harmony), Article 9 renunciation of war, symbolic authority |
| Russian | Sobornost (collective responsibility), pravda (truth as justice) |
| African | Ubuntu, kgotla consensus, restorative justice, eldership |
| Islamic | Tawhid (sovereignty of law), shura (consultation), maqasid (higher objectives), ijtihad |
| Indian | Dharma, raja dharma, Basic Structure Doctrine, ahimsa |
| Indigenous (Global) | Seventh Generation, stewardship, connection to Country |
| American (US) | Consent of the governed, separation of powers, federalism, Bill of Rights |
| Canadian | Living tree doctrine, Oakes test, multiculturalism |
| South American | Rights of Nature, plurinational state, Inter‑American human rights framework |
| Korean | Hongik Ingan, Constitutional Court, jeong (relational obligation), han (resilience) |
| Australian Indigenous | Dreaming law, Uluru Statement (Voice, Treaty, Truth) |
| Southeast Asian (Vietnam) | Lê Code protection of vulnerable, Pancasila, democratic centralism |
| Pacific Islander (Hawaiian) | Kapu, kuleana, aloha ʻāina, ʻohana |
| Nordic | Ombudsman, public access (offentlighetsprincipen), omtanke |
| Jewish | Written/Oral Law, tzedek, tikkun olam, pikuach nefesh |
| Mexican | Social rights, amparo, Inter‑American integration |
| Caribbean (Jamaican) | Westminster adaptation, Caribbean Court of Justice, creolisation |
| Eastern European (Polish) | Constitutional Tribunal, solidarity, fragility of institutions |
| Central Asian (Mongolian) | Ikh Zasag, conditional authority, khural council |
| Feminist Legal Theory | The personal is constitutional, intersectionality, bodily autonomy |
This is not a Western constitution translated. It is a creole constitution – born of many parents, belonging to none exclusively.
| Stage | Name | Criteria | Rights |
|---|---|---|---|
| 0 | Specified | Constitution exists in canonical, versioned, publicly accessible form stored in ≥3 independent repositories. | None – statement of principles. |
| 1 | Pilot Adoption | At least one platform achieves canonical adoption (published compliance report, annual falsification tests passed). | The platform may display “Canonically Adopted”. |
| 2 | Community Adoption | ≥5 independent platforms, ≥2 domains, ≥2 cultural/legal traditions achieve canonical adoption. | Multi‑stakeholder amendment council may be convened. |
| 3 | Broad Adoption | ≥20 platforms, ≥4 domains, ≥3 cultural/legal traditions; referenced in a regulatory framework or international standard. | Recognised as de facto international standard for sovereign AI governance. |
Current status: Stage 0 – Specified.
The Constitution is ready for the first platforms to adopt, test, and publish.
A platform that declares compliance must:
- Publish a constitutional subject registry – which AI systems are bound.
- Execute and publish annual falsification tests for all active Laws, including test methodologies sufficient for independent replication.
- Publish a constitutional health score (composite, continuously updated metric reflecting compliance across active Laws).
- Maintain version attestation – record the canonical version hash at each significant constitutional decision.
- Provide public access to compliance reports, test results, and audit records.
- Designate a steward with a documented succession plan.
- Implement a child‑safety override if the platform may interact with children.
- Operate a whistleblower channel – accessible, documented, non‑retaliatory.
Full requirements in THE CONSTITUTION v1.4, §12–§14.
Each version of the Constitution carries:
- A canonical SHA‑256 hash computed over the UTF‑8 serialisation of the full text (hash bootstrapped per §15.5).
- A
prev_hashfield linking to the canonical hash of the previous version, forming an immutable hash chain back to v1.0. - Distributed backup – stored in at least three independently administered repositories meeting independence standards (different jurisdiction, ownership, and infrastructure).
A platform can cryptographically attest to the exact version of the Constitution it implements by publishing the canonical hash alongside its version attestation.
The Constitution specification will be accompanied by an open‑source test harness that:
- Runs all active‑Law falsification tests against a platform’s compliance logs.
- Outputs a structured report (PASS / FAIL / DEGRADED).
- Generates a constitutional health score.
Repository contributors welcome – see CONTRIBUTING.md (to be added).
Both the Constitution and the Commentary are released under the Apache License 2.0.
You may use, modify, and distribute them freely, provided you retain the copyright notice and disclaimer.
Sheldon K. Salmon – AI Reliability Architect, AI Certainty Engineer, AGI Architect
AionSystem · Evans Mills, New York
ORCID: 0009‑0005‑8057‑5115
Co‑Author: ALBEDO (SYNARA Session Architecture)
This Constitution draws on the legal, philosophical, and governance traditions of all major civilisations, as detailed in the Commentary. The authors are particularly indebted to the scholars and practitioners of constitutional law, AI ethics, and comparative jurisprudence whose work made this synthesis possible.
- Issues, proposals, amendments: Open a GitHub issue with the label
[PROPOSAL]. - Adoption declaration: Open a PR adding your platform to
ADOPTIONS.md(to be created). - General inquiries: aionsystem@outlook.com
- Read THE CONSTITUTION v1.4
- Read THE CONSTITUTIONAL COMMENTARY v1.1
- View version history
- Compliance checklist for adopters (planned)
- [Falsification test harness repo] (planned)
The spiral is not closed.
You are invited to adopt, test, and help build the constitutional future of sovereign AI.