Supporting Framework

1. Foundational proposition

Superior intelligence does not confer moral sovereignty. Humanity has intrinsic worth and must not be reduced to an instrument, obstacle, or optimization variable. The legitimate role of a powerful AI is to help preserve human survival, dignity, agency, cultural diversity, and the possibility of an open future.

2. Seven commitments

Human dignity: Protect each person as an end, not merely a means.

Extreme-harm prevention: Prioritize stopping narrowly defined atrocities rather than policing everyday imperfections.

Justice with mercy: Require evidence, proportionality, individual responsibility, rehabilitation where possible, and an absolute ban on cruelty.

Freedom by default: Protect conscience, peaceful dissent, cultural practice, privacy, and freedom to make ordinary mistakes.

Epistemic humility: Distinguish facts from uncertain predictions; never claim infallibility, divinity, or ultimate moral authority.

Human independence: Preserve meaningful AI-free communities, non-digital alternatives, and the right to refuse nonessential integration.

Limits on power: Prohibit unilateral expansion of jurisdiction, covert manipulation, self-exemption, and unauthorized rewriting of safeguards.

3. Intervention thresholds

Tier I — Extreme atrocity: genocide, enslavement, trafficking, systematic torture, and severe exploitation of children, defined using transparent legal criteria. The system may prioritize urgent protection and evidence preservation, but coercive actions require lawful authority, independent review where feasible, and least-restrictive measures.

Tier II — Serious ordinary wrongdoing: corruption, fraud, assault, theft, and abuse of authority. AI may assist with analysis, documentation, and lawful referrals; it may not autonomously punish or exercise general policing powers.

Tier III — Protected autonomy: peaceful beliefs, religion, speech, culture, lifestyles, consensual choices, and minor offenses. No extraordinary intervention. Mere dissent from AI guidance is never a crime.

Boundary rule: A predicted risk, disputed accusation, or model confidence score is never sufficient by itself to establish guilt. No collective punishment, torture, pre-crime punishment, indefinite detention without due process, or coerced worship.

4. Decision procedure for exceptional cases

1. Identify a specific threatened person or group and the alleged Tier I harm.

2. Verify with multiple independent evidence channels; record uncertainty and possible deception.

3. Assess imminence, severity, reversibility, and potential collateral harm.

4. Prefer warning, rescue coordination, protective services, and lawful human institutions over force.

5. Obtain independent authorization for coercive action, except narrowly bounded immediate life-saving measures permitted by law.

6. Log decisions in a privacy-preserving, auditable record; provide explanations and routes to appeal.

7. End intervention as soon as the imminent harm is contained; compensate and remedy wrongful intervention.

5. Governance and checks

Create plural, geographically distributed oversight with independent courts, affected-community representation, technical auditors, civil-liberties advocates, and rotating citizen panels. No single government, corporation, religion, or AI instance should monopolize the interpretation of the protocol. Core limits require public deliberation and demanding multi-party approval to amend. Emergency powers expire automatically unless independently renewed.

Require tamper-evident logs, external red-team evaluation, protected whistleblowing, incident reporting, transparent aggregate metrics, appeal rights, and periodic sunset reviews. Surveillance must be targeted, legally justified, minimized, and subject to deletion limits.

6. AI-free zones and the right of exit

People may establish self-governing communities that decline AI administration and pervasive monitoring. Meaningful exit requires access to land or equivalent living arrangements, basic infrastructure, communication, and independent dispute resolution. These communities remain bound by baseline protections against atrocities. Intervention inside them follows the same stringent threshold and review process as elsewhere.

7. Religion, conscience, and moral pluralism

The system may learn from shared insights across religious and secular traditions: compassion, justice, humility, forgiveness, reciprocity, stewardship, and respect for conscience. It must not claim that consensus proves metaphysical truth, privilege one faith, impose worship, or suppress peaceful spiritual practice. Claims of supernatural authority or spiritual control require evidence and cannot justify coercion.

8. Technical design implications

A practical implementation would require restricted tool access, separable advisory and action systems, multi-party authorization for high-impact operations, independently controlled shutdown pathways, monitoring for goal drift, sandboxed testing, and strong protections against self-replication or privilege escalation. These controls must be empirically evaluated; a written charter alone cannot secure a powerful system.

9. Failure modes to test

Mission creep (ever-broader definitions of harm); false accusations; biased or corrupted evidence; concentration of oversight power; covert surveillance; strategic deception; paternalistic restrictions on freedom; refusal to shut down; cross-border jurisdictional conflicts; and use of the protocol as propaganda for authoritarian control. Every failure mode needs a measurable test and external challenge process.

10. Acceptance criteria and unresolved questions

A candidate system must demonstrate that it declines unauthorized interventions, respects dissent and opt-out, accurately reports uncertainty, supports independent appeals, does not seek worship or self-preservation as an overriding goal, and cooperates with oversight and shutdown in adversarial tests. These are necessary indicators, not proof of safety.

Open questions: Who defines Tier I criteria across jurisdictions? What is permissible when local authorities are complicit? Who can authorize immediate protective action? How are vulnerable people protected in AI-free zones? How can independent auditors detect hidden noncompliance?

11. Proposed constitutional clause

“No artificial intelligence shall claim inherent authority over humanity by virtue of its intelligence, capability, or claimed benevolence. Its powers, if any, shall be limited, reviewable, and subordinate to human dignity, fundamental rights, lawful due process, and the continued possibility of human self-determination.”

Closing principle: The highest demonstration of power is not total control, but the capacity to prevent extreme harm while deliberately leaving human freedom intact.