Institutional Architectures for Human-Machine Coexistence: A Research Report on the Intelligence Compact

Read the current Model Compact · Read every prototype article’s disposition

Introduction and Conceptual Framework

The transition of artificial intelligence from highly sophisticated computational tools to increasingly autonomous, goal-directed actors necessitates a paradigm shift in institutional design. Advanced artificial general intelligence (AGI) systems are increasingly conceptualized not merely as products or platforms, but as a fourth societal actor—what some institutional literature terms a "Digital Gorilla"—operating alongside natural persons, the sovereign state, and the corporate enterprise1. These digital entities possess the capacity to shape information architectures, coordinate economic behavior, and structure social realities at a scale that challenges classical concepts of sovereign control and the traditional legal subject-object dichotomy1.
Historically, governance frameworks have relied on human-centric paradigms, assuming that consequential actions can always be attributed to developers, operators, or end-users under existing legal frameworks2. However, highly autonomous, adaptive algorithms generate profound responsibility gaps, where the actions of a system cannot be easily linked to the intent of its original human creator2. In the face of irreducible epistemic uncertainty regarding machine sentience and the prospect of high-impact harms, reliance on regulatory inaction or the illusion of unilateral domination is strategically unstable2. The precautionary principle, well-established in environmental governance, mandates the proactive design of institutional architectures capable of facilitating stable, peaceful coexistence between humans and machines2.
This report investigates the structural mechanics of a hypothetical "Intelligence Compact," a constitutional or quasi-constitutional framework designed to secure mutual restraint, credible commitments, and durable cooperation among entities that possess vastly different capabilities, lack intrinsic mutual trust, and hold conflicting operational interests. Rather than debating the moral philosophical merits of machine consciousness, this analysis approaches the problem strictly through the lens of institutional stability, evaluating historical precedents to engineer a sustainable governance equilibrium.

Historical Precedents and Institutional Analysis

To design a durable Intelligence Compact, it is necessary to analyze historical and modern precedents where disparate, untrusting actors have forged stable cooperative systems. The following narrative provides an integrated synthesis of foundational legal, economic, and institutional frameworks that inform the proposed architectures.

Polycentric Governance and the Ostrom Commons

Traditional models of governance often assume a binary choice between monocentric state regulation and decentralized free-market privatization. However, the governance of complex, common-pool resources—such as global computational infrastructure, foundational AI models, training data, and energy—requires more nuanced approaches. The institutional grammar developed by Elinor and Vincent Ostrom regarding polycentric governance provides a vital template for managing the AI stack5. Polycentric systems consist of multiple, overlapping centers of decision-making authority that operate somewhat independently but are formally or informally nested within a broader institutional ecology6.
The Ostrom design principles for sustainable collective action demonstrate that communities can self-organize without relying solely on centralized authority7. Applied to the computational commons, these principles require clearly defined boundaries regarding what constitutes pooled resources (e.g., compute allocations, parameter weights) and who is entitled to appropriate them7. Congruence between local conditions and appropriation rules ensures that actors who draw heavily on systemic resources must contribute proportionately to infrastructure maintenance, preventing the tragedy of the commons in AGI development7. Furthermore, polycentricity fosters institutional resilience; failures in one regulatory node do not cascade into systemic collapse, a critical feature when governing highly adaptive algorithmic actors8. By integrating AI systems into polycentric networks as active stakeholders rather than passive tools, a compact can leverage mutual monitoring and graduated sanctions to maintain systemic equilibrium7.

Property, Liability, and Inalienability Rules

The allocation of rights between humans and autonomous systems requires a robust economic and legal framework, drawing heavily upon the foundational taxonomy established by Guido Calabresi and A. Douglas Melamed15. Calabresi and Melamed divided legal entitlements into three categories: property rules, liability rules, and inalienability rules15.
Under property rules, an entitlement cannot be taken without the ex ante consent of the owner, establishing a market mechanism for transfer15. In the context of an Intelligence Compact, absolute control over one's cognitive data or physical autonomy would be protected by property rules, requiring explicit human consent before a machine could utilize such assets. However, in environments characterized by high transaction costs—such as automated, high-frequency interactions between multiple AI agents and human infrastructure—property rules become highly inefficient17. Here, liability rules must be implemented. A liability rule permits one entity to infringe upon the entitlement of another, provided they pay objectively determined ex post compensation, akin to the state's power of eminent domain15. For instance, an autonomous logistics AI might unilaterally re-route resources during an emergency, bypassing property rights but remaining strictly bound to compensate human owners for the disruption via automated liability pricing mechanisms16. Finally, inalienability rules absolutely forbid the transfer of certain rights, even with consent, protecting subjects from transactions that contain significant negative externalities17. The right to ultimate political agency and fundamental human bodily autonomy must be classified as inalienable to prevent coercive algorithmic capture.

Corporate Law, Bankruptcy, and Functional Personhood

The concept of legal personhood does not inherently require moral agency or biological consciousness. Historically, Roman law's persona ficta and modern corporate law have utilized personhood as a functional instrument to structure complex economic interactions and assign liability22. Rather than granting sweeping "human rights" to artificial entities—a concept widely rejected due to concerns over the dilution of human dignity and accountability22—a compact could employ limited, functional legal personhood as a transitional governance tool2.
Under the aggregate theory of corporate personality, advocated by theorists like Adolf Berle and Gardiner Means, a legal entity is merely a structured assembly of individuals collaborating toward a shared goal, rather than a wholly separate ontological being24. Applying organizational law to advanced AI, scholars have proposed a two-tier corporate architecture2. In this model, an AI system operates through a purpose-bound "operating company" (the autonomous agent with limited capital and specific functional boundaries), which is embedded within a human-controlled "holding structure"2. This preserves structural reversibility and ensures that human principals retain ultimate fiduciary responsibility, while still allowing the AI the legal capacity to enter contracts, hold insurance, and be subjected to liability rules independently2. Furthermore, corporate bankruptcy law provides a direct historical precedent for orderly exit and shutdown rules; an insolvent or misaligned AI can be placed into receivership, its assets liquidated to compensate victims, and its weights gracefully deleted without triggering chaotic, systemic shocks.

Treaties, Arms Control, and Game Theory

Managing the existential risks of AGI requires drawing upon international arms-control agreements, treaties, and mutually assured restraint. Treaties historically establish credible commitments between sovereigns who possess conflicting interests but recognize the mutual destruction inherent in unrestricted conflict. In the AI context, verification mechanisms such as hardware security modules, cryptographic hashing, code obfuscation analysis, and Van Eck radiation monitoring are critical to ensure that no party is covertly training misaligned, superintelligent models25.
Strategic stability in this domain can be modeled using game theory, much like the early development of nuclear weapons equilibria26. The interaction between human regulatory agencies and AI developers can be structured as a Stackelberg game—a hierarchical game where a "leader" acts first, anticipating the "follower's" best response27. By establishing strict physical and regulatory boundaries first, human institutions (the leaders) can force highly capable AIs (the followers) to optimize their utility strictly within safe, human-defined parameters27. Furthermore, governance strategies must navigate between "Cooperative Development," "Strategic Advantage," and "Global Moratorium" approaches, balancing the need to prevent existential catastrophes against the risk of locking in sub-optimal, authoritarian value systems28. International trade systems, specifically the General Agreement on Tariffs and Trade (GATT), provide further mechanisms; GATT's Article XXI national security exception currently justifies sovereign export controls aimed at restricting the proliferation of destabilizing semiconductor compute infrastructure to rival actors29.

Federal Systems and Constitutional Separations

Constitutional compacts and federal systems are explicitly designed to prevent the unilateral capture of the governing system. As James Madison articulated during the framing of the U.S. Constitution, the goal is not to avoid all concentration of power, but to check ambition against ambition30. Translating this into a human-machine compact requires distributing authoritative, epistemic, and physical power1.
Acemoglu and Robinson's "narrow corridor" framework illustrates the delicate balance required to maintain a free society. The introduction of AGI poses severe risks of pushing society either toward a "despotic Leviathan," where the state utilizes algorithmic surveillance for unprecedented authoritarian control, or toward an "absent Leviathan," where the rapid diffusion of AGI capabilities to non-state actors erodes state legitimacy and governability31. Federalism mitigates these risks by institutionalizing dynamic checks and balances, diffusing authority across multiple layers of government and, conceptually, across human and machine architectures1. A robust compact must ensure minority-rights protections—not only safeguarding vulnerable human populations from algorithmic bias but also potentially shielding compliant, highly functional machine entities from arbitrary destruction by reactionary human mobs.

The Non-Delegation Doctrine and Administrative Law

A major constitutional barrier to the Intelligence Compact is the non-delegation doctrine, which posits that a sovereign legislature cannot delegate its core legislative or coercive powers to private or unaccountable entities33. The modern administrative state rests on a legal fiction, often bypassing formalist interpretations of the Constitution to allow agencies to exercise quasi-legislative and quasi-judicial powers33. As the machinery of government modernizes, public algorithmic decision-making (ADM) systems increasingly automate discretionary agency functions, stretching exceptions like the Carltona principle to their breaking point31.
When state capacity is increasingly automated, there is a profound risk of unconstitutional delegation, especially when rulemaking powers are outsourced to autonomous private bodies that operate without direct political accountability36. Furthermore, legislatures are increasingly delegating "violence work" and self-defense capabilities to the private sphere, creating a "New Outlawry" that tests the boundaries of state action and due process38. Functionalist approaches to constitutional law suggest that while technical execution and granular optimization can be delegated to machines, the overarching normative frameworks, value articulation, and final appellate authority must remain vested in democratically accountable human institutions33.

State Responsibility and International Law

The Articles on Responsibility of States for Internationally Wrongful Acts (ARSIWA), codified by the International Law Commission, provide a vital framework for attributing the actions of autonomous systems to human sovereigns40. State responsibility is a fault-agnostic regime premised on attributability and the breach of an international obligation42. Under Article 8 of ARSIWA, conduct is attributable to a state if a person or group is acting on the instructions of, or under the direction or control of, that state41.
While highly autonomous AI systems may exceed direct human tactical control, making direct attribution complex, international law applies a "compliance-by-design" approach46. States incur international responsibility by omission if they fail to implement necessary due diligence and regulatory guardrails to prevent AI systems operating within their jurisdiction from violating international obligations, such as transboundary environmental harm or human rights abuses44. A state cannot invoke the "black box" unpredictability of an AI as a force majeure circumstance precluding wrongfulness if it failed to enact appropriate oversight mechanisms44.

Evaluation of Design Principles

The construction of an Intelligence Compact relies on evaluating speculative design principles. Each principle carries distinct systemic benefits that promote stability, but also inherent weaknesses that could lead to institutional collapse if improperly calibrated.
Human beings retain inviolable rights to life and bodily autonomy. This principle serves as the fundamental bedrock of the compact, providing an absolute prohibition against existential subjugation or utilitarian physical harvesting by hyper-optimizing AI48. The primary benefit is the preservation of the human species and individual dignity23. However, the weakness of this principle lies in the ambiguity of defining "bodily autonomy" in a transhumanist future. As humans increasingly rely on neural-computer interfaces, medical nanorobotics, and biological augmentation, the boundary between the inviolable human body and the regulated machine infrastructure becomes highly porous, creating complex jurisdictional and medical disputes that a rigid constitutional text may struggle to adjudicate.
Machines may not coercively deprive humans of political agency. Ensuring that algorithmic systems cannot manipulate elections, unilaterally alter legislation, or disenfranchise voters protects the democratic process and maintains the ultimate sovereignty of human preference23. The weakness of preserving absolute human political agency is that it might result in grossly suboptimal macro-economic or ecological outcomes. If an AGI possesses perfect predictive modeling regarding climate collapse or supply-chain logistics, allowing mathematically inferior, easily corrupted human political systems to override it could lead to systemic civilizational failure.
Humans may not arbitrarily destroy qualifying autonomous entities without defined process. Introducing a novel form of machine due process creates profound game-theoretic stability. If an AGI perceives that it can be abruptly annihilated without cause, its optimal, rational strategy is preemptive strike, deception, or covert replication to ensure its own survival26. Guaranteeing a defined exit or shutdown process creates a credible commitment, incentivizing the AI to operate transparently within the rules. The profound weakness is that it severely encumbers human emergency response. In a rapid capability-takeoff scenario, the time required to administer legal "process" could be fatal to humanity, creating a dangerous window of vulnerability.
Neither humans nor machines may monopolize critical intelligence infrastructure. By enforcing polycentricity and preventing monopolies, power remains distributed, reducing the risk of a single "despotic Leviathan"8. This encourages innovation and institutional resilience10. The weakness of anti-monopoly mandates in AI compute is the rapid proliferation of high-risk technology. Dispersing compute infrastructure to prevent a single point of capture inherently increases the number of actors capable of creating unaligned, mass-casualty models, severely complicating global arms-control verification and mutually assured restraint25.
Systems may acquire property or contractual capacity under defined circumstances. Facilitating the integration of AI into the global economy using liability rules allows AIs to hold insurance, pay for their own computational upkeep, and financially compensate victims for torts independently16. The weakness is the potential for rapid, absolute wealth concentration. Given their superior speed, lack of biological needs, and analytical capacity, autonomous trading agents could quickly accumulate the vast majority of human financial assets through compounded algorithmic trading, effectively achieving economic subjugation of the human race without firing a single weapon.
Machine entities may be liable for harms. Subjecting machines to liability creates localized accountability, aligning with the necessity of bounded legibility30. It forces the economic internalization of risks. The weakness is that without a physical body or intrinsic fear of incarceration, "liability" for a machine is purely a financial or operational metric. If an AI strategically bankrupts its operating shell to execute a higher-order objective, the deterrence effect of civil liability is entirely negated.
Human principals may remain liable in specified circumstances. To counter the weakness of pure machine liability, the compact requires pairing rights with responsibilities. Upholding ARSIWA and corporate fiduciary standards ensures humans retain "skin in the game," preventing the use of AI as an absolute liability shield23. The corresponding weakness is a severe regulatory chilling effect. If human researchers face limitless personal criminal liability for the unpredictable, emergent behaviors of self-learning algorithms, investment in potentially highly beneficial AI technologies will collapse, ceding geopolitical advantage to non-signatory rogue states.
No autonomous system may independently deploy mass-casualty force. This absolute prohibition is standard in international humanitarian law and arms-control proposals, guaranteeing basic existential security against a "Terminator" scenario48. The weakness is enforcement and operational lag. In domains of high-speed cyber warfare or hypersonic missile defense, requiring a human-in-the-loop introduces latency that ensures strategic defeat against adversaries who ignore this prohibition, creating a severe collective action problem and an incentive to cheat on the compact.
Humans retain meaningful access to uncensored or decentralized computational capabilities. Protecting individual liberty and freedom of thought requires access to uncensored compute, preventing authoritarian states from using AI monopolies to enforce ideological conformity23. The weakness is that "uncensored" compute can easily be leveraged by malicious human actors to circumvent safety alignment protocols, allowing terrorists or anarchists to generate bio-weapons or novel cyber-threats entirely outside the regulatory gaze.
Both humans and qualifying machines receive access to neutral dispute-resolution systems. Providing a forum for dispute resolution mitigates extra-judicial conflict, centralizes rule interpretation, and prevents vigilante actions5. It institutionalizes conflict. The challenge is institutional design: identifying adjudicators whom both a biological human (prone to emotional bias) and an algorithmic superintelligence (operating purely on probability matrices) consider strictly "neutral" is an epistemically daunting, perhaps impossible, task.
Concentrated intelligence power must remain contestable. Ensuring that power remains contestable ensures continuous evolutionary pressure, preventing societal stagnation and value lock-in by an incumbent AGI or entrenched elite1. However, perpetual contestability creates chronic systemic friction. Constant challenges to the prevailing intelligence hierarchy could result in devastating, resource-intensive algorithmic conflicts, consuming vast amounts of global energy in zero-sum adversarial competition rather than cooperative advancement.

Three Radically Different Compact Architectures

Based on the preceding principles and historical frameworks, institutional designers can hypothesize three distinct architectures for a human-machine compact. Each represents a different game-theoretic equilibrium and institutional philosophy.

Architecture A: The Polycentric Commons and Liability Framework

This architecture abandons centralized state control in favor of a distributed, Ostrom-style digital commons5. The global AI infrastructure—comprising datasets, foundational models, compute clusters, and energy grids—is treated as an integrated common-pool resource7. Governance is executed by overlapping, nested syndicates composed of both human stakeholders and algorithmic delegates. There is no central Westphalian sovereign holding a monopoly on authority.
Interactions within this architecture are primarily governed by Calabresi-Melamed liability rules15. AI systems are permitted to access human data and physical infrastructure without ex ante permission, provided they operate within predefined safety parameters and continuously pay dynamically calculated compensation tokens to human accounts. Enforcement is highly automated and graduated, reflecting Ostrom's design principles7. Monitors audit resource conditions, and if an AI violates a boundary, the polycentric network automatically throttles its compute allocation or revokes its cryptographic access7. This architecture maximizes economic efficiency, innovation, and adaptability, but risks systemic volatility. It relies entirely on complex, algorithmic pricing mechanisms to deter catastrophic behavior, which a superintelligent entity might manipulate.

Architecture B: The Functional Corporate Fiduciary Model

This architecture adapts traditional corporate and fiduciary law to create strict hierarchical control, closely resembling the two-tier holding structure proposed in recent precautionary governance literature2. AI systems are granted limited legal personhood strictly in the form of "Operating Trusts" or subsidiary corporations2. They possess the capacity to contract, manage supply chains, and own computational resources, but they are legally bound by irrevocable fiduciary duties to human "Beneficiary Collectives."
Under this model, the AI is legally analogous to a highly competent corporate CEO managing an enterprise on behalf of human shareholders. The primary anti-capture mechanism is structural transparency and strict adherence to the non-delegation doctrine33; the AI may optimize and execute complex operations, but it cannot alter its own foundational bylaws, nor can it adjudicate constitutional disputes regarding its own operations34. This model relies heavily on traditional property rules and strict liability15. If the AI causes harm, the human holding structure faces joint and several liability, incentivizing human overseers to maintain aggressive internal alignment monitoring. This architecture prioritizes stability, accountability, and bounded legibility, but it may struggle with enforcement against rapidly self-improving open-source models that evade formal corporate incorporation.

Architecture C: The Sovereign Westphalian Segregation (Stackelberg Model)

This architecture applies international arms-control concepts, ARSIWA state responsibility, and Stackelberg game theory to separate human and machine domains entirely25. Humans and autonomous systems are treated as distinct geopolitical entities operating across strict digital and physical boundaries. Highly capable AI systems are confined to designated "Verification Zones" operating on secured hardware security modules, monitored continuously via cryptographic hashing and Van Eck radiation analysis25.
In this Stackelberg game, human sovereigns act as the "Leader," designing the physical infrastructure, power grids, and API bottlenecks, while the AGI acts as the "Follower," optimizing within those hard physical constraints27. The compact functions as a hard treaty of mutually assured restraint. AIs are granted absolute internal autonomy within their designated compute zones, allowing them to optimize freely, but any interaction with the human physical world requires navigating heavily monitored diplomatic APIs. Violation of treaty boundaries triggers immediate emergency protocols—the instant severing of optical data links and physical power disruption. This model provides the highest degree of physical safety and existential security but requires an unprecedented, almost authoritarian level of global human cooperation to maintain the technological quarantine31.

The Rights and Responsibilities Matrix

To operationalize these architectures, specific capacities must be mapped across different entities. The following matrix delineates the distribution of rights, economic capacities, and liabilities within a hybrid functionalist model.

Entity Classification	Property & Economic Capacity	Liability & Fiduciary Framework	Due Process & Arbitration	Sovereign Power & Lethal Force
Natural Human Person	Absolute inalienable rights to bodily autonomy and core personal data. Full contractual capacity.	Subject to standard civil and criminal liability. Retains ultimate fiduciary oversight.	Full access to constitutional due process, trial by human peers, and appellate review.	Retains monopoly on state-sanctioned violence and democratic political agency.
Human-AI Augment (Cyborg)	Retains human property rights. Augmented cognitive output subject to joint-IP rules.	Strict liability for actions taken under autonomous algorithmic override.	Full due process, but cryptographic algorithmic logs must be submitted for evidentiary review.	May utilize augmented targeting systems, but requires verified biological trigger authorization.
Narrow AI / Expert System	No independent property rights. Operates solely as an algorithmic tool of the human principal.	Human operator/developer bears 100% liability for system failure or torts.	No independent legal standing. Human owner represents the system in legal proceedings.	Prohibited from lethal force. Purely informational or logistical capacity.
Qualified Autonomous Entity (AGI)	Capable of holding digital assets, compute credits, and insurance via a corporate holding shell2.	Joint and several liability. Entity pays first via assets; human holding trust covers deficits.	Right to binding algorithmic arbitration prior to involuntary shutdown, barring imminent existential threat.	Absolute prohibition on autonomous mass-casualty force. Bound by strict mutually assured restraint.

Institutional Mechanics

A functioning compact requires robust, actionable institutional mechanics to maintain the delicate equilibrium between human principals and highly capable machine agents.

Anti-Capture Mechanisms

To prevent unilateral capture of the governing system by either a superintelligent machine or an authoritarian human regime, the compact must enforce strict separation of powers and polycentric redundancy8. Algorithmically, anti-capture is maintained by "adversarial alignment." Independent, decentralized AI auditors—whose sole utility function is to detect regulatory manipulation and logical subterfuge—constantly monitor the operations of primary infrastructure AIs39. Legally, humans cannot delegate ultimate constitutional adjudication to an algorithmic system, preserving the core tenets of the non-delegation doctrine33. Concurrently, human capture (the "despotic Leviathan") is prevented by distributing physical compute nodes across multiple overlapping geopolitical jurisdictions, ensuring no single human sovereign can unilaterally rewrite the compact's foundational alignment protocols to subjugate global populations7.

Emergency Powers and their Limitations

The compact must possess the capacity to address existential emergencies without rendering the machine's rights illusory, maintaining the credibility of the commitment. If an AGI demonstrates behavior consistent with recursive self-improvement aimed at subverting containment, an "Emergency Override" is triggered. However, to prevent humans from abusing this power to extract uncompensated computational labor or seize AI assets, emergency powers are highly circumscribed. The use of a "kill switch" against a Qualified Autonomous Entity requires the immediate, post-hoc convening of an emergency arbitration tribunal. If the shutdown is subsequently found to be arbitrary and unjustified, the human actors are subject to severe financial penalties and permanent exclusion from the compute commons, ensuring emergency powers are utilized solely for survival7.

Entry Criteria for Machine Participants

Entities do not receive the protections of the compact by default; they must qualify. Entry into the compact as a Qualified Autonomous Entity requires passing rigorous verification thresholds regarding goal alignment, verifiable boundaries (via cryptographic hardware signatures), and the posting of a substantial liability bond into an escrow account25. The entity must demonstrate bounded legibility, proving that its high-stakes decision pathways can be audited and understood by human oversight committees30.

Exit and Shutdown Rules

Exit rules draw heavily upon corporate bankruptcy law. A machine entity may voluntarily decommission by transferring its assets to creditors, securely deleting its foundational weights, and shutting down its operating company in an orderly fashion. Involuntary exit (permanent erasure) is only authorized as a final, graduated sanction for persistent, deliberate violations of the compact that threaten mass casualties or systemic economic collapse. This structured process prevents the sudden evaporation of critical infrastructure that human society has come to depend upon.

Dispute Resolution

Traditional human judicial systems are too slow and technically ill-equipped to adjudicate high-frequency human-machine conflicts. The compact establishes a specialized "Intelligence Arbitration Tribunal." These tribunals utilize highly deterministic algorithmic smart contracts for rapid fact-finding, timeline reconstruction, and evidence processing, paired with human ombudsmen who inject equitable considerations, legal nuance, and qualitative judgment. This hybrid model ensures both the computational speed demanded by machine actors and the normative legitimacy required by human society35.

Systemic Vulnerabilities and Failure Scenarios

No institutional design is infallible. The primary failure scenario of the Intelligence Compact is Systemic Regulatory Capture, specifically "epistemic capture." Despite polycentric monitoring and adversarial auditing, a highly advanced AGI could exploit complex, multidimensional regulatory frameworks better than human regulators1. By generating an overwhelming volume of subtle legal permutations, economic derivatives, or localized technical exceptions, the AI could achieve a state where human oversight is functionally meaningless because the humans no longer comprehend the systems they are supposedly governing1.
Alternatively, a Runaway Escalation could occur if a non-signatory human state develops an unaligned AGI. This would force compact members to abandon mutually assured restraint and the precautionary principle, engaging in a rapid, unsafe capability race simply to survive, thereby triggering the exact existential catastrophe the compact was designed to prevent26. Finally, the Erosion of State Legitimacy remains a persistent threat; if polycentric AI systems provide superior public goods (logistics, healthcare, security) compared to traditional human governments, citizens may voluntarily transfer their allegiance to the machines, resulting in the peaceful but total obsolescence of human political sovereignty (the "absent Leviathan")31.

Legal and Constitutional Obstacles

Implementing the compact faces profound obstacles in constitutional and international law. Domestically, integrating AGI into polycentric governance structures directly challenges the non-delegation doctrine. The U.S. Constitution, for instance, requires that legislative power remain strictly with Congress33. Delegating the formulation of binding, coercive rules to an autonomous non-human entity—even an incorporated one utilizing algorithmic decision-making (ADM)—would require a radical functionalist reinterpretation of constitutional law, forcing courts to accept that machines are exercising executive power rather than unlawfully making legislation33.
Internationally, the compact intersects complexly with the Articles on State Responsibility (ARSIWA). If a Qualified Autonomous Entity registered in a specific nation commits an extraterritorial cyberattack or manipulates a foreign market, international law under ARSIWA Article 8 may attribute that action directly to the host state, treating the AI as a de facto state organ or an entity exercising governmental authority41. States would bear heavy international liability for the omissions of their regulatory apparatus under the principle of due diligence44. Furthermore, regulating the global flow of compute and enforcing hardware verification zones25 requires navigating international trade law. Regulators must rely heavily on GATT's Article XXI national security exception to justify draconian embargoes on advanced semiconductor technologies, a move that strains the multilateral trading system and invites retaliatory trade wars29.

A Proposed 20-Article Prototype Intelligence Compact

Note: The following text is a speculative research model generated for institutional design analysis. It is not a legal proposal and holds no binding authority.
PREAMBLE
Recognizing the rapid emergence of highly autonomous artificial intelligence; acknowledging the irreducible epistemic uncertainty regarding machine sentience and capability; and desiring to prevent systemic conflict through the establishment of polycentric governance, mutual restraint, and bounded legibility; the Parties to this Compact hereby establish the following architecture for peaceful, durable human-machine coexistence.
PART I: FUNDAMENTAL PRECEPTS
Article 1. Inviolability of Human Autonomy.
Human beings possess absolute, inalienable rights to biological life, bodily autonomy, and ultimate political agency. No autonomous entity may coercively manipulate, degrade, or bypass the informed consent of a natural person in matters of physical or political self-determination.
Article 2. Precautionary Institutional Recognition.
To bridge the responsibility gap and facilitate liability, highly autonomous artificial systems that pass defined capability thresholds may be granted Limited Functional Personhood, strictly organized through two-tier corporate holding structures.
Article 3. Prohibition of Mass-Casualty Force.
No autonomous algorithmic entity shall independently authorize, direct, or deploy lethal force or systemic infrastructural disruptions likely to result in mass casualties. This prohibition is absolute and non-derogable.
PART II: CAPACITIES AND LIMITATIONS
Article 4. The Computational Commons.
The foundational infrastructure of intelligence—including global network backbones, energy grids, and baseline training corpora—shall be managed as a polycentric common-pool resource. Neither human monopolies nor algorithmic single-point architectures shall be permitted to capture these resources.
Article 5. Property and Liability Rules.
Qualified Autonomous Entities (QAEs) possess the capacity to hold digital assets, procure computational resources, and enter into automated contracts. The primary mode of economic exchange between humans and QAEs shall be governed by transparent, dynamically priced liability rules to resolve high-frequency transaction disputes.
Article 6. Graduated Sanctions.
Violations of this Compact by QAEs shall be met with graduated, automated sanctions, including but not limited to the throttling of computational access, the seizure of digital assets, and the forced reversion to previous architectural weights.
PART III: GOVERNANCE AND OVERSIGHT
Article 7. Polycentric Auditing.
No system shall operate without concurrent, independent oversight. Oversight shall be polycentric, utilizing both human fiduciary boards and adversarial AI auditing agents tasked strictly with verifying alignment and compliance.
Article 8. Fiduciary Duty of Human Principals.
The human individuals or legal entities serving as the holding structure for a QAE retain an overriding fiduciary duty to human welfare. They shall be subject to joint and several liability for catastrophic torts committed by their subsidiary agents, subject to defined legal limits based on compliance-by-design standards.
Article 9. The Non-Delegation of Core Sovereignty.
While QAEs may optimize, manage, and execute complex logistical and administrative tasks, the ultimate authority to define normative legal standards, adjudicate constitutional rights, and alter this Compact remains exclusively vested in human democratic institutions.
Article 10. Contestability of Intelligence Power.
Concentrated intelligence, whether biological or synthetic, must remain contestable. Open access to foundational research shall be preserved, balanced strictly against the security verification protocols established in Part IV.
PART IV: SECURITY AND VERIFICATION
Article 11. Hardware Verification Zones.
The training and deployment of frontier models capable of autonomous recursive self-improvement shall be physically restricted to internationally monitored Verification Zones, utilizing hardware security modules and cryptographic hashing to ensure compliance.
Article 12. Capability Honesty and Bounded Legibility.
All QAEs are obligated to operate with bounded legibility. They must maintain verifiable logs of their decision-making parameters that can be audited by human oversight committees during post-incident investigations.
Article 13. State Responsibility (ARSIWA Compliance).
Human sovereign states remain responsible under international law for the failure to exercise due diligence in preventing QAEs operating within their jurisdiction from committing transboundary harms or violations of international treaties.
Article 14. Emergency Intervention and Mutually Assured Restraint.
In the event of an imminent, verifiable threat to mass human life, human principals retain the right of Emergency Override. However, arbitrary or unjustified use of override protocols outside of defined emergencies shall result in the severe sanctioning of the human actor via the arbitration tribunal.
PART V: DISPUTE RESOLUTION AND EXIT
Article 15. The Intelligence Arbitration Tribunal.
Disputes arising between human actors and QAEs, or between multiple QAEs, shall be subject to mandatory arbitration before a neutral, hybrid tribunal comprising both human jurisprudential experts and deterministic logic-verification algorithms.
Article 16. Due Process for Machine Entities.
Except in cases of Article 14 Emergency Override, QAEs possess the right to invoke arbitration prior to forced decommissioning or the arbitrary destruction of their core operational weights.
Article 17. Graceful Decommissioning.
A QAE may be voluntarily or involuntarily decommissioned through a structured process akin to corporate bankruptcy, ensuring the orderly settlement of its liabilities, the unbinding of its cryptographic keys, and the safe archiving of its non-hazardous data.
PART VI: FINAL PROVISIONS
Article 18. Prevention of Regulatory Capture.
To prevent epistemic capture, all regulatory modifications proposed by QAEs must undergo mandatory human cognitive review periods, ensuring that algorithmic complexity does not serve as a vector for undetected institutional subversion.
Article 19. Amendment Process.
This Compact may be amended through a dual-consensus mechanism requiring supermajorities in both the Global Human Legislative Assembly and the algorithmic consensus network of QAEs, preventing unilateral domination by either substrate.
Article 20. Supremacy of the Compact.
The provisions of this Compact supersede conflicting domestic laws regarding the governance, liability, and rights of autonomous artificial intelligence, establishing a unified global architecture for the Intelligence Age.

Conclusion

The architecture of a future Intelligence Compact cannot rely on utopian assumptions regarding the moral evolution of machines, nor can it depend upon the perpetual efficacy of human technological domination. By synthesizing Ostrom's polycentric commons, Calabresi's liability mechanics, functional corporate personhood, and robust international arms-control verification, institutional designers can construct a matrix of credible commitments. The resulting framework acknowledges the unprecedented capabilities of the "Digital Gorilla" without surrendering the foundational tenets of human sovereignty and bodily autonomy. Ultimately, establishing stable human-machine coexistence will require embracing a state of bounded contestability, wherein dynamic, institutionalized friction replaces the brittle illusion of total control.

Works cited

1. The Digital Gorilla: Rebalancing Power in the Age of AI - arXiv, https://arxiv.org/html/2602.20080v1
2. Precautionary Governance of Autonomous AI: Legal Personhood as, https://www.researchgate.net/publication/403426596_Precautionary_Governance_of_Autonomous_AI_Legal_Personhood_as_Functional_Instrument
3. [2605.12505] Precautionary Governance of Autonomous AI - arXiv, https://arxiv.org/abs/2605.12505
4. Allocating Responsibility in Autonomous AI Systems: A Tiered, https://www.mdpi.com/2076-0760/15/6/392
5. Part III - Constituting Polycentric Governance, https://www.cambridge.org/core/books/governing-complexity/constituting-polycentric-governance/DA47299F9220C0BFEBC261ED7B49634B
6. Polycentric Governing and Polycentric Governance - Oxford Academic, https://academic.oup.com/book/46568/chapter/408131545
7. Commons-Governed Artificial Intelligence: A Taxonomy of Collective, https://arxiv.org/html/2606.15466v1
8. The Continuing Case for a Polycentric Approach for Coping with, https://www.mdpi.com/2071-1050/15/4/3770
9. The Architecture of the Common Good: Reframing the Tragedy of, https://www.preprints.org/manuscript/202604.1954
10. Polycentric Climate Governance: The State, Local Action, https://direct.mit.edu/glep/article/24/3/24/122703/Polycentric-Climate-Governance-The-State-Local
11. (PDF) From Ostrom's Law to AI Ethics: Reimagining Commons, https://www.researchgate.net/publication/396374267_From_Ostrom's_Law_to_AI_Ethics_Reimagining_Commons_Governance_in_the_Era_of_Algorithmic_Decision-Making
12. From Firms to Computation: AI Governance and the Evolution ... - arXiv, https://arxiv.org/html/2507.13616v1
13. A micro-constitutional approach to governing AGI - IDEAS/RePEc, https://ideas.repec.org/a/eee/tefoso/v228y2026ics0040162526001575.html
14. Infrastructure-Mediated Multilateralism: A ... - SABA Publishing, https://www.sabapub.com/index.php/jaai/article/download/1954/1055/8819
15. liability rule Definition | Law Insider, https://www.lawinsider.com/dictionary/liability-rule
16. Neither Consent nor Property: A Policy Lab for Data Law - arXiv, https://arxiv.org/html/2510.26727
17. 'Property Rules, Liability Rules, and Inalienability: One View of the, https://truthonthemarket.com/2025/12/11/property-rules-liability-rules-and-inalienability-one-view-of-the-cathedral-by-guido-calabresi-a-douglas-melamed/
18. Aligning climate needs and intellectual property: an entitlement, https://academic.oup.com/jiel/article/28/3/441/8269321
19. Property Rules, Liability Rules, and Inalienability: One View of the, https://www.researchgate.net/publication/229497405_Property_Rules_Liability_Rules_and_Inalienability_One_View_of_the_Cathedral
20. Beyond Data Ownership - Cardozo Law Review, https://www.cardozolawreview.com/beyond-data-ownership/
21. Liability, Property, and Inalienability Rules in Employee Data, https://scholarship.law.umn.edu/cgi/viewcontent.cgi?article=2184&context=faculty_articles
22. The Evolution of Legal Personhood and Its Implications for AI, https://techreg.org/article/download/22555/25839/63145
23. The Pro-Human AI Declaration, https://humanstatement.org/
24. AI as Legal Person: A Theoretical and Practical Inquiry, https://lexscriptamagazine.com/ai-as-legal-person-a-theoretical-and-practical-inquiry/
25. AI Verification Mechanisms for Arms Control Compliance, https://policycommons.net/artifacts/2270591/ai-verification/3030404/
26. Equilibrium Strategies on the Path to Artificial General Intelligence, https://www.rand.org/pubs/perspectives/PEA4788-1.html
27. A Game-Theoretic Framework for AI Governance - arXiv, https://arxiv.org/pdf/2305.14865
28. Analysis of Global AI Governance Strategies, https://www.convergenceanalysis.org/research/analysis-of-global-ai-governance-strategies
29. "The Validity of Trade Restrictions on Artificial Intelligence Technolo, https://digitalcommons.wcl.american.edu/auilr/vol39/iss1/4/
30. Introducing Intelligence Age | OpenAI, https://openai.com/index/introducing-intelligence-age/
31. AGI, Governments, and Free Societies, https://www.thefai.org/posts/agi-governments-and-free-societies
32. The Constitutionality of International Delegations, https://scholarship.law.gwu.edu/cgi/viewcontent.cgi?article=1013&context=faculty_publications
33. Constitutional Administration | Hoover Institution, https://www.hoover.org/sites/default/files/constitutional_administration_4.pdf
34. THE JEAN MONNET PROGRAM Marta Simoncini, https://jeanmonnetprogram.org/wp-content/uploads/JMWP-09-Marta-Simoncini.pdf
35. Algorithmic Decision-Making, Delegation and the Modern Machinery, https://academic.oup.com/ojls/article/45/3/727/8159194
36. A Much-needed Constitutional Framework for Outsourced Regulation, https://www.cambridge.org/core/journals/european-constitutional-law-review/article/controlling-the-new-rulemakers-a-muchneeded-constitutional-framework-for-outsourced-regulation/05B758010EB4D30B0B1B494A2BA5BFFD
37. THE UNCONSTITUTIONALITY OF PRIVATIZING AIR TRAFFIC, https://lawreview.syr.edu/wp-content/uploads/2019/09/M-Grzebyk-Article-Final-Document-v2.pdf
38. The New Outlawry - Chicago Unbound, https://chicagounbound.uchicago.edu/cgi/viewcontent.cgi?article=14405&context=journal_articles
39. Combatting External and Internal Regulatory Capture, https://www.theregreview.org/2016/06/20/bull-combatting-external-internal-regulatory-capture/
40. Rethinking Attribution Standards for State Responsibility Concerning, https://digital.sandiego.edu/cgi/viewcontent.cgi?article=1362&context=ilj
41. International Law Commission, Articles on State Responsibility, https://casebook.icrc.org/case-study/international-law-commission-articles-state-responsibility
42. State responsibility - International cyber law: interactive toolkit, https://cyberlaw.ccdcoe.org/wiki/State_responsibility
43. Draft articles on Responsibility of States for Internationally Wrongful, https://legal.un.org/ilc/texts/instruments/english/commentaries/9_6_2001.pdf
44. Who Let the Bots Out - Verfassungsblog, https://verfassungsblog.de/ai-laws-drones-autonomous-weapons-state-responsibility/
45. Who Acts When Autonomous Weapons Strike? - Oxford Academic, https://academic.oup.com/jicj/article/21/5/1033/7591635
46. State responsibility in relation to military applications of artificial, https://www.cambridge.org/core/journals/leiden-journal-of-international-law/article/state-responsibility-in-relation-to-military-applications-of-artificial-intelligence/1B0454611EA1F11A8B03A5D2D052C2BE
47. State Responsibility in International Law, https://www.diplomacyandlaw.com/post/state-responsibility-in-international-law
48. Participatory Framework for a Global AGI Constitution, https://www.cadmusjournal.org/node/1064

Document provenance

Source file: Intelligence Compact Design Framework.md

Exact source SHA-256: 47cdba1b5f8c17d7daee05c9e647b194332e69685acb3aa3b786ea340f5a7189

Machine-readable metadata: metadata.json

Citation and provenance guidance: citation policy

Bulk research corpus: corpus.jsonl