Bridging Intelligence
with Decision Sovereignty
through Accountable Continuity

Architecting the transition from stalled AI investment to
enterprise value that compounds as context evolves.

Decision Governance:

The Biggest Mistake in AI Ethics: We Protect Humans While We Stop Developing Them.

Technological power does not confer the right to govern. Disarming AI is not enough. The world needs decision owners, Mission Continuity and institutions designed to improve human capability and independence.

Andre Rekhtine | September 23, 2026

2026 AI wave is repeating both 1973 Cybersyn and 1999 DotCom

The world may be solving the wrong AI crisis

The human may still be in the loop. That may be the problem.

For years, the defining fear of artificial intelligence has been that machines might escape human control. It is a spectacular possibility, dramatic enough for headlines, political hearings and warnings about the end of civilization. Yet power does not always announce its transfer with an emergency. More often, it arrives quietly, disguised as assistance.

The machine does not need to seize the decision. It can begin by summarizing the evidence, identifying the risks and ranking the available options. It can gradually decide which facts deserve attention, which uncertainties can be ignored and which outcomes appear reasonable. A human still reviews the recommendation, clicks approve, signs the order and carries the legal liability. Everything remains reassuringly human, except for the context that produced the decision.

A person who can approve an action but cannot reconstruct the context from which it emerged does not control the system. That person merely completes its workflow.

This is the quieter AI crisis taking shape beneath the public argument about safety. The danger is not only that machines may remove humans from the loop. It is that institutions may preserve the appearance of human control after the human has lost the information, authority and independent judgment required to own the decision.

That possibility sounds frightening, but it is not inevitable. AI can extend human judgment rather than replace it. It can expose overlooked evidence, test entrenched assumptions and give ordinary institutions capabilities once reserved for the largest corporations and intelligence services. The question is not whether humans should use AI. They already do, and in many cases they should. The question is what remains human after they do.

DECISIVE QUESTION

Who will own the context, authority and consequences when machine intelligence becomes institutional action?

How ownership disappears while authority appears to remain

Context Ownership rarely vanishes in a single transfer. It weakens through two mutually reinforcing mechanisms.

The first is perceived authority. AI need not issue an order or own the final action. It can acquire practical authority earlier by framing the decision context, borrowing trust from the institution or interface through which it speaks, and making one interpretation appear more complete and actionable than the evidence supports. Formal decision rights remain human while the reality within which those rights are exercised has already been assembled elsewhere.

The second is cognitive atrophy. When AI repeatedly performs source review, synthesis, comparison, exception handling and interpretation, independent human reasoning becomes progressively less necessary in daily work. The immediate output may improve while the institution loses the practice through which people learn to reconstruct context, challenge authoritative conclusions and own consequential decisions.

Perceived authority captures today's decision. Cognitive atrophy captures the institution's ability to challenge tomorrow's decision. The organization becomes more productive while consuming its future capacity to govern.

Fear is not only a warning. It is a competition for the role of saviour

The political narrative that AI may kill everyone turns a set of specific institutional design failures into an undifferentiated Existential Threat. Once that conversion occurs, fear begins to perform two political functions at the same time.

First, it justifies accelerated national investment because a rival must not arrive first. Second, it justifies centralized approval, surveillance and control because ordinary institutions are said to be incapable of governing the danger. Acceleration becomes necessary because the rival may win. Centralization becomes necessary because the technology may escape. In both cases, responsibility migrates away from named human decision makers.

All major centres of power have reason to sustain some version of this fear. Each is competing not merely to protect humanity, but to become its preferred saviour, and each arrives with a different blueprint for the future.

The United States offers technological leadership. China offers coordinated controllability. Europe offers rights-based regulation. Britain offers stewardship and public service. International bodies offer negotiated standards. Frontier laboratories offer technical expertise and access to the systems that only they “fully understand”.

This does not mean that every warning is manufactured. The risks may be genuine while the competition to define the remedy is equally genuine. A centre of power can sincerely fear a danger and still gain authority by presenting itself as the institution uniquely qualified to save society from it.

The pattern has appeared during earlier crises involving public health, climate, energy and national security. Those crises are materially different and should not be collapsed into one category. The recurring governance mechanism is subtler: existential threats create a demand for salvation. Institutions then compete not merely to protect humanity, but to become its legitimate saviour. Each arrives with a preferred vision of the future and a claim that only its path can safely lead there. As the demand for salvation grows, so does the demand for coordination, visibility and exceptional authority. Exceptional authority can then outlive the emergency that justified it and extend into financial, social and institutional life far beyond the original crisis.

The question is not whether the threat is real or false. The question is whether those seeking the authority to save humanity remain subject to named decision ownership, lawful authority, contestable context and consequences when that authority is exercised beyond its legitimate bounds.

Every metaphor brings an institution with it

AI is increasingly compared with nuclear technology. The analogy is understandable because both can alter the balance of power, create risks beyond national borders and generate pressure for international guardrails. Yet every metaphor carries a ready-made institution inside it.

Calling AI the next nuclear problem does more than describe the scale of danger. It begins to select the response: secrecy, strategic rivalry, restricted access, expert control, international inspection and exceptional state authority. Recent discussion of United States and Chinese guardrails has already invoked the logic of nuclear non-proliferation, even while acknowledging that AI is easier to distribute and that neither open nor closed models are safe merely because of their architecture. [1]

The nuclear analogy may help leaders recognize the stakes, but it can also hide the nature of the problem. Nuclear material can be counted, guarded and physically contained. AI lives inside software, communications, institutions and ordinary workflows. Its most consequential effect may not be a single explosion. It may be the gradual relocation of judgment into systems that no person fully understands and no institution completely owns.

This is why the competing doctrines should be examined before a new global architecture is chosen. Each doctrine offers a real strength. Each also seeks legitimacy as the answer to humanity's fear.

Orwell, Huxley and the dual meaning of “both”

The emerging international debate recalls two dystopian visions that are often treated as alternatives. George Orwell imagined submission produced by fear, surveillance and coercion. Aldous Huxley imagined something more comfortable: people surrendering autonomy in exchange for convenience, stimulation and relief from the burden of independent judgment.

One future frightens the human into obedience. The other makes obedience feel like a service. The emerging AI order may not force humanity to choose between them. It may combine them.

One architecture tells people to accept expanding supervision because AI is dangerous. The other encourages them to surrender judgment because AI is useful, effortless and increasingly difficult to live without. Orwell's citizen obeys because resistance is punished. Huxley's citizen gradually loses the ability to explain why resistance would be desirable.

The comparison is not an accusation that AI is inherently totalitarian. The same systems can distribute knowledge, make expertise more accessible and help individuals challenge institutional monopolies. AI can strengthen independent judgment when it exposes assumptions, invites challenge and leaves the human with the information and authority required to alter the outcome. The problem begins when convenience becomes dependence and supervision becomes the permanent price of access.

At Dreamforce 2026, Jensen Huang rejected the choice between rapid development and product safety as false: “Speed, and safe products... it's a false choice. You can definitely have both at the same time.” He argued that companies should move quickly but pause when they lose confidence in control or product safety. [2]

In its engineering context, this is compelling. Safety should be designed into progress rather than purchased through paralysis. Mature engineering disciplines become safer through testing, redundancy, accountability and operational control. Society should demand speed and safety.

But context changes the meaning of “both.” In a political context, the same formula can become maximum technological acceleration surrounded by maximum supervisory capacity. The technology must move faster because a rival may arrive first, while control must expand because the technology is moving too fast. Each becomes the justification for the other.

Huxley makes the system indispensable. Orwell makes its supervision legitimate. The darkest interpretation of “both” is therefore not speed with safety, but speed for the builders and control over everyone else.

It does not have to end that way. The real question is who defines safety, who acquires authority in its name and who owns the decision when speed and safety come into conflict.

What Cybersyn actually attempted

Cybersyn is often remembered through its operations room, futuristic chairs and real-time economic data. Its deeper ambition was organizational. Stafford Beer and the Chilean team were trying to create a viable relationship among distributed operations, exception signals, human interpretation and authority to act.

Raul Espejo later emphasized that the clients needed the organizational model, not merely the software. Cole Cioran's public treatment of that history restores the point that the technology was never the whole system. The system depended on relationships, feedback, escalation and responsibility.

This is why Cybersyn belongs in the AI discussion now. Agentic systems are becoming vastly more capable at observing, generating and transmitting signals. Yet the institutional question remains recognizable: who is authorized to interpret the signal, who can intervene, and who owns the consequence when the operating picture changes?

Sovereignty reveals the same defect at institutional scale

Robert Stoneman's briefing on digital sovereignty describes a world in which governments and institutions can lose practical control through vendor dependency, foreign jurisdiction and rapidly changing geopolitical conditions. The visible issue is sovereignty, but the underlying governance problem is continuity of legitimate decision authority.

A state or enterprise may know where data resides and still be unable to act without another party's permission. It may possess infrastructure and still lack access to the evidence, services or operational context needed to decide. Residency is necessary. It is not sufficient.

Decision sovereignty is the capacity to form a decision from legitimate context, authorize it, enact it without impermissible dependency, preserve accountability, and continue deciding as conditions change.

Owning the walls is not the same as owning what happens inside them.

The AI Factory makes the next step unavoidable

A recent financial-services blueprint co-authored by John Ratzan, Paul Barrett and Satish Mariyappa describes the production-side shift with unusual clarity. The Agentic AI Factory moves the enterprise from disconnected pilots toward repeatable AI operations, and from assistants that summarize toward decision systems that retrieve context, apply business rules, coordinate specialized capabilities and participate in core workflows.

The significance of the blueprint is not any particular vendor configuration. It is the recognition that infrastructure, orchestration, data, governance, workflow integration, regulatory alignment and organizational change must operate as one production system.

Industrialized intelligence cannot be created by scaling models or infrastructure in isolation.

This is an important movement beyond AI theatre.

It also exposes the next operating-model question.

A decision system can be traceable, explainable, monitored and technically constrained.

The institution must still determine who holds the integrated context when the system’s output becomes consequential action. The factory can industrialize intelligence. It cannot, by itself, industrialize legitimate human authority or continuing accountability.

Industrializing intelligence and preserving the human capacity to govern it may be the same operating-model problem viewed from opposite ends.

The apprenticeship contradiction

Ratzan's separate warning about the erosion of apprenticeship makes the tension more consequential. Coding, legal research, financial analysis and first-draft work were not merely low-level tasks. They were part of the repeated practice through which professional judgment was formed.

If AI removes that work, institutions may scale decision systems while weakening the human pipeline expected to govern them.

This is not only a reskilling problem. It is a continuity problem.

The enterprise must redesign how judgment, authority and accountability are developed and transferred. Otherwise it may gain abundant machine intelligence while losing the human capacity to recognize when context has changed, when a technically legitimate answer has become institutionally wrong, or when an automated action must be stopped.

A skill is not a structure. Human discernment matters, but discernment without authority is a well-informed bystander. Authority without context is merely power. Accountability without continuity becomes blame after the fact.

Cynefin diagnoses the changing domain

Dave Snowden's Cynefin framework provides a disciplined way to distinguish different causal conditions. Clear situations permit established practice.

Complicated situations require expertise and analysis. Complex situations require probes through which patterns can emerge. Chaotic situations require action to create enough stability for sensemaking to resume.

AI transformation is frequently managed as complicated when it is actually complex. Organizations expect a roadmap, select a platform and attempt to scale a predefined operating model. Yet the deployment itself changes workflows, incentives, authority, evidence and the meaning of the decisions being automated.

Cynefin identifies the nature of the situation and protects leaders from applying the wrong mode of action. It does not, by itself, identify the person who owns the evolving interpretation or remains accountable after the intervention changes the system.

The competing doctrines

The United States: sovereignty through acceleration

The American doctrine begins with a strategic warning: AI is not merely another industry waiting to be regulated. It is a foundational capability that will shape productivity, defence, intelligence, infrastructure and geopolitical power. A country that slows itself unilaterally may not create a safer world. It may simply surrender the future to a rival that does not slow down.

In this model, investment becomes security, scale becomes protection and innovation becomes sovereignty. The state does not disappear. It protects the technological frontier, enforces existing law, defends critical infrastructure and helps maintain national leadership. Frontier laboratories, cloud providers, chip manufacturers, defence institutions and capital markets become parts of an increasingly integrated national capability. [3]

This is not an irrational position. The United States cannot preserve economic or military independence by abandoning technologies that others will continue to develop. Its strength is velocity, capital formation and an unusual ability to convert research into deployed capability.

Its weakness is that national strength does not automatically preserve human agency. A country may possess the world's leading models while public servants, employees and citizens become dependent on systems whose context they cannot inspect. National sovereignty is not institutional sovereignty, and institutional sovereignty is not decision sovereignty. A country can win the AI race while people acting in its name lose the ability to understand, contest and own the decisions produced by its systems.

China: sovereignty through coordinated capacity

The Chinese doctrine begins from a different premise. AI development, industrial policy, infrastructure and public governance are treated as elements of a coordinated national project. An expanding open-weight ecosystem offers countries and organizations an alternative to complete dependence on closed Western frontier laboratories. Models that can be downloaded, studied and modified may distribute technical capability more broadly. [1]

This model contains a legitimate challenge to technological dependency. Open access can reduce the power of a small group of vendors, while coordinated investment can build infrastructure at a scale fragmented markets may struggle to achieve. China can therefore reject Western existential fear while supporting state-centred international AI governance. Resisting restrictions prevents the existing technological hierarchy from being frozen in place, while multilateral governance raises the role of governments relative to private laboratories.

The underlying promise is that coordinated public capacity can prevent private technology companies from becoming sovereign powers. Its corresponding risk is that state coordination can absorb local judgment. A system can be nationally independent, technically accessible and strategically coordinated while leaving the individual decision maker with little freedom to challenge the context supplied from above.

State sovereignty may protect a society from foreign dependency. It does not guarantee that the people inside the state own consequential decisions.

The European Union: sovereignty through rules

The European doctrine begins with a third proposition: technological power can be constrained through law, standards, rights and access to the market. Where the United States emphasizes leadership and China emphasizes coordinated capacity, the European Union seeks to make legitimate AI conditional on compliance.

Its comparative advantage is not the largest frontier laboratory or the most unified state-directed technology stack. It is the ability to convert political principles into requirements for companies seeking access to its market. Risk classification, transparency, documentation, human oversight and fundamental rights are intended to prevent technological capability from outrunning democratic authorization. [4]

That is not bureaucratic irrelevance. Law is one of the few instruments capable of binding power before harm occurs. Without enforceable rules, human rights can become optional product features.

Yet compliance can create a particularly convincing illusion of control. A system may satisfy every documentation requirement while presenting a human decision maker with a context already filtered, ranked and framed by others. An audit trail can prove that a person clicked approve. It cannot prove that the person owned the judgment.

Europe is therefore structurally vulnerable to the darker interpretation of “both.” It cannot afford to abandon the AI race, but it is inclined to make acceleration acceptable by surrounding it with a denser architecture of certification, monitoring and institutional supervision. Each element can be defensible in isolation while the combined system moves decision sovereignty upward and away from the person expected to bear the consequences.

Britain: sovereignty through stewardship and security

Britain occupies a related but distinct position. Its emerging doctrine combines national-security institutions, AI safety capability, defence-industrial partnerships and an ambition to shape international standards. Its public language places key technological decisions under elected government while presenting the United Kingdom as an honest broker capable of coordinating global risk management. [5]

This model has real strengths. Britain possesses security institutions, scientific credibility, diplomatic reach and a constitutional tradition capable of framing power as stewardship rather than possession. It can convene actors that neither markets nor regulators can bring together on their own.

Its risk is similar to Europe's but more executive in character. Stewardship can become supervisory authority, and emergency preparedness can grow into permanent visibility over increasingly integrated systems. Britain, too, may seek to preserve Huxleyan convenience while adding Orwellian assurance around it, not because it embraces either dystopia, but because speed and control together appear more governable than either alone.

FOUR DOCTRINES

America accelerates. China coordinates. Europe regulates. Britain supervises and convenes. Each protects something real, but none yet guarantees human ownership of the consequential decision.

The Crown sees farther

The British Crown adds something that the four administrative doctrines cannot supply by themselves: a constitutional horizon longer than an electoral cycle and a moral language larger than national competition.

In September 2026, King Charles III convened leaders from frontier laboratories, technology companies, government and ethical institutions to discuss AI in service of the public good and the need for meaningful human control. The importance of this intervention is not that the Crown should govern technical systems. It is that the Crown can legitimize a question markets and bureaucracies tend to subordinate: what is technological power for, and in whose service should it operate? [6]

The Crown sees farther because it is not required to reduce the future to the next product cycle, election or regulatory milestone. It can speak for continuity, duty, communities, future generations and the natural world. It can remind the system that capability remains legitimate only while it serves a purpose beyond its own expansion.

For Canada, this is not foreign symbolism. The Canadian Crown is legally distinct from the Crown in right of the United Kingdom, and the King does not direct Canadian policy. Yet the Sovereign remains embedded in Canada's constitutional order. The Crown can therefore supply one moral-authority node in a Canadian system that already separates enduring constitutional legitimacy from the temporary authority of government.

Moral convening, however, is not an operating model. The Crown can ask whether humanity remains principal rather than subject. It cannot by itself identify the Context Owner, align authority with accountability or preserve the next generation of decision owners. For that, another source of moral direction is needed.

The Vatican illuminates the direction

The Vatican changes the subject of AI governance from the machine to the human person. Magnifica Humanitas asks not only whether technology remains controlled, regulated or useful, but what kind of human being and society the technology is producing. Its principles of dignity, conscience, subsidiarity and human development provide the moral direction that technical safety frameworks lack. [7]

Subsidiarity places decisions as close as possible to the people they affect, while higher authorities support rather than replace human freedom and responsibility. Its account of work treats work not merely as income or output, but as a setting in which people acquire competence, responsibility, relationships and maturity.

This matters because AI can increase output while removing the work through which judgment is formed. Junior research, comparison, drafting, ambiguity and correction are not simply costs around expertise. They are the apprenticeship through which future experts are produced. A system can protect a person from error while depriving that person of the experience required to become capable.

The Crown names legitimate service and continuity. The Vatican points toward the human person, conscience, subsidiarity and development. Together they reveal the missing principle: governance must not merely prevent AI from harming humanity. It must preserve and improve humanity's capacity to judge, choose and remain independent.

The fantasy of the universal kill switch

Nothing exposes the gap between political imagination and operational reality more clearly than the demand for an AI kill switch. The image is irresistible. Somewhere there is a secure room and a red button. If AI becomes dangerous, a responsible authority presses it, the systems stop, humanity exhales and civilization resumes on Monday morning.

It is a comforting picture. At the scale implied by the political rhetoric, it is also a fantasy.

A kill switch can be meaningful inside a bounded system. An operator can terminate an agent session, revoke credentials, disconnect network access, restrict permissions, throttle an API or suspend a service. These are necessary engineering controls, and targeted intervention may be safer than shutting down the underlying model. [8]

A universal switch is different. There is no single machine to stop and no single authority controlling every model, copy, provider, local deployment and open-weight system. AI already exists across competing laboratories, government systems, private infrastructure and devices. A mechanism capable of reaching all of them would require an extraordinary architecture of surveillance and command, and would itself become a premier target for cyberattack, political abuse or catastrophic error. [8][9]

The more important question is almost never asked: what happens on the Monday after the switch is pressed? By the time AI is embedded in logistics, cybersecurity, software development, banking, health administration and public services, switching it off will not restore a pristine human baseline. The old processes may no longer exist at their former capacity. Institutional knowledge may have migrated into automated workflows. Staffing models may assume permanent AI availability. Humans may no longer possess the practiced skills or current context required to reconstruct interrupted operations manually.

The danger is not only that the switch might fail. The danger is that it might work exactly as designed.

A universal kill switch capable of disabling an AI-dependent economy would cease to be a safety mechanism. It would become a controlled-demolition mechanism without a recovery plan. The proper alternative is layered containment, graceful degradation, narrow permissions, redundancy and Mission Continuity. The aim is not to destroy the surrounding system in order to prove that humans remain in control. It is to let humans interrupt dangerous actions while preserving the society the technology has come to support.

MISSION CONTINUITY

A kill switch preserves the ability to stop the machine. Mission Continuity preserves the human capability to continue after it stops.

Canada has already learned the price of authority without timely correction

Canada is unusually prepared to understand this problem because it possesses a system of dual moral authority and has recently paid a visible price for the exercise of emergency power beyond its lawful threshold.

Its constitutional inheritance combines the Crown's ethic of continuity, duty and restrained authority with a Catholic and post-Catholic tradition of human dignity, community and development that remains especially visible through Quebec. The Quiet Revolution transferred many social functions from Catholic institutions to the secular state, but it did not erase the developmental question. It changed the institutional carrier responsible for answering it. [10]

Canada has also seen how quickly an emergency narrative can move coercive authority into private infrastructure. In response to the 2022 Freedom Convoy, the federal government invoked the Emergencies Act and issued the Emergency Economic Measures Order. Financial institutions were required to identify designated persons, disclose information and cease dealings with affected property. Private banks became operational extensions of emergency enforcement, affecting not only direct participants but donors and supporters connected through financial networks. [11][12]

On January 16, 2026, the Federal Court of Appeal held in Canada (Attorney General) v. Canadian Civil Liberties Association, 2026 FCA 6 that the invocation was unreasonable and ultra vires. The Court found that the statutory threshold had not been met and upheld findings involving freedom of expression and protection against unreasonable search and seizure. [11]

The point here is not to relitigate every claim surrounding the convoy. It is to identify the governance lesson. Emergency authority acted immediately. Private infrastructure operationalized it immediately. The constitutional correction arrived almost four years later, when the immediate effects could no longer be reversed by a judgment.

This is precisely the temporal asymmetry that AI can intensify. Machine-speed enforcement can make a legally contestable decision operational before Parliament, courts or the public can reconstruct its context. If accountability arrives years after a bank account, licence, service or livelihood has been disabled, accountability remains essential but no longer restores the lost decision moment.

Canada therefore knows the price of protection without a named owner, bounded authority and real-time contestability. It also knows why institutional checks matter. That experience should not disqualify Canada from AI leadership. It may be exactly what qualifies Canada to lead a more mature model.

Why Canada is ready to lead the real third path

Canada should not choose between American innovation velocity, Chinese coordinated control and European regulatory protection. Britain's stewardship adds a valuable constitutional dimension, but stewardship alone does not produce an operating model. The Crown sees farther. The Vatican illuminates the direction. Canadian law reveals the cost of overreach. Canada can turn those insights into governable practice.

Canada’s opportunity is not to join the competition to become humanity’s next protector. It is to change the architecture of protection itself: to ensure that no protector can convert necessity into unlimited authority, accountability never exceeds the actual power to govern a decision, and no AI deployment is considered successful if it weakens the human capability and independence required to own the next consequential decision.

That is the real third path. It begins with the reconstructed Zeroth Law and becomes operational through Double-Bound AI Ethics.

The reconstructed Zeroth Law

Asimov's Zeroth Law placed humanity above obedience and robotic self-preservation by prohibiting a robot from harming humanity or, through inaction, allowing humanity to come to harm. Its lasting insight was systemic: the consequence for the whole cannot be ignored. Its limitation was equally systemic. A prohibition supplies a boundary, not a developmental mission.

The reconstructed law changes the sign. Avoiding harm is the minimum condition, not the final purpose.

RECONSTRUCTED ZEROTH LAW

Every consequential decision must advance the mission of improving the capabilities and independence of humanity and the human person. The goals and values of every nested institution must serve the same purpose.

This principle prevents a local objective from becoming self-legitimating. A company cannot justify permanent human dependency merely by increasing productivity. A regulator cannot justify unlimited control merely by reducing risk. A government cannot protect civic order by destroying civic agency. A model cannot be called aligned if its deployment gradually eliminates the people capable of challenging it.

A single empty office chair drawn in gold dotted outline on white, labelled Context Ownership Void, standing for the missing accountable owner of the decision.

Dual Ethics: two moral actors, not one compliance layer

Most AI governance assumes that ethics can be placed inside the technical system. The model should be aligned, prohibited actions should be blocked, rules should be encoded and outputs should be monitored. This is necessary, but insufficient.

A technically constrained system can still serve an institution pursuing the wrong objective. A compliant model can operate inside emergency authority that has exceeded its mandate. A safe recommendation can arise from incomplete context. An ethical machine can be used by an institution that has transferred responsibility into procedure.

Dual Ethics recognizes two moral actors. The first is the technical system and the organization operating it. They must respect explicit boundaries, identify uncertainty, preserve traceability and avoid actions outside authorized scope. The second is the human decision owner, who must retain the capability and duty to question the objective, reconstruct the context, challenge the recommendation and refuse an action that remains technically permissible but has become morally indefensible.

The machine must not cross its boundary. The human must not surrender judgment at theirs.

A human approval button is therefore not enough. The human must possess access to the relevant context, authority proportional to accountability, the practical ability to interrupt the action and protection from retaliation when that interruption is exercised responsibly. Dual Ethics does not place the human ceremonially in the loop. It makes the human capable of governing the loop.

Double-Bound AI Ethics

The first bound constrains technological and institutional power. Privacy, security, fairness, transparency, testing, access controls, data provenance, incident response, independent evaluation, appeal rights and restrictions on irreversible automation remain essential.

The second bound develops human ownership. Every significant deployment must be evaluated by its developmental yield. Does the system strengthen judgment or replace its practice? Does it create more people capable of understanding the decision context? Does responsibility grow with authority? Can the institution continue deciding if the model or provider disappears? Is a new generation learning to recognize exceptions, challenge assumptions and stop an action?

The first bound prevents capability from escaping acceptable limits. The second prevents the human being from degrading safely inside those limits. Together they define Human-Development-Centric AI.

The AI Stability Board can become the international platform

Canada does not need to invent an international platform from nothing. The proposed AI Stability Board offers the institutional opening.

In September 2026, Canadian Prime Minister Mark Carney proposed a Technology Stability Board modelled on the Financial Stability Board. Paul Samson of the Centre for International Governance Innovation described how a coalition of willing countries could begin with common model evaluations, system documentation, incident reporting and a small coordinating secretariat. The proposal develops an earlier Digital Stability Board concept intended to bridge the governance gap between globally operating platforms and fragmented national institutions. [13][14]

A stability board could provide the coordination and shared-evidence layer missing from international AI governance. Yet stability remains a protective objective. The board would remain incomplete if it measured model capability and incidents without asking whether AI leaves people and institutions more capable of reconstructing context, owning decisions and operating independently of the systems they govern.

Financial stability protects a system from collapse. AI stability must do more. It must prevent humanity from becoming stable inside managed dependency.

Canada can lead by giving the board a dual mandate.

Track One: protective stability

• Coordinate common model evaluations, system documentation and serious-incident reporting.

• Create cross-border comparability for capability, misuse, containment and operational risk.

• Define layered emergency controls instead of relying on a mythical universal switch.

• Require clear authority, evidence thresholds, expiry conditions and review for emergency intervention.

• Test whether critical services can degrade safely when models, vendors or networks fail.

Track Two: developmental stability

• Identify Context Ownership Voids before consequential deployment.

• Measure whether institutions retain independent access to primary evidence and the ability to reconstruct decisions.

• Test whether accountability is matched by usable authority and stop capability.

• Monitor cognitive atrophy and the removal of formation-bearing work.

• Trace Mission Lineage from the immediate objective through every nested institution to the improvement of human capability and independence.

• Assess whether another prepared owner can inherit the decision contour when the current owner leaves.

• Require Mission Continuity for high-consequence deployments.

The board should not become a world government for AI. Its legitimacy would depend on remaining a coordination, evidence and standards platform that strengthens local decision ownership rather than absorbing it. Subsidiarity must apply internationally as well as institutionally: higher levels support, compare and intervene only where lower levels cannot responsibly manage the risk.

The Monday Morning Test

Every high-consequence AI deployment should pass a simple but unforgiving test: if the system stops on Friday night, which consequential decisions must still be makeable on Monday morning, by whom, using what context, under whose authority and through what degraded operating capability?

Technical recovery is not enough. The institution must preserve purpose, living context, decision rights, consequence ownership and enough practiced human capability to continue its legitimate mission without the system.

MONDAY MORNING TEST

If the AI stops on Friday night, who can still own the consequential decision on Monday morning, with what context, authority and degraded operating capability?

Seven public requirements

Name the consequential decision. Governance begins with the choice and consequence, not the system inventory.

Name the Context Owner. Someone must maintain the integrated and changing context from which the decision acquires meaning.

Name the Decision Owner. A capable human must possess legitimate authority to approve, refuse, stop, redirect and reopen.

Connect ownership to consequences. Accountability must never exceed actual authority, and correction must remain part of ownership.

Trace Mission Lineage. The immediate purpose and every nested mission must remain coherent with improving human capability and independence.

Preserve Mission Continuity. Every high-consequence deployment must maintain a degraded operating mode in which legitimate human decisions remain makeable after the system becomes unavailable or untrusted.

Preserve the path to the next owner. AI-supported work must retain the apprenticeship through which another person learns to reconstruct context and inherit the decision contour.

The final lesson

The world does not need another centre of power competing to save humanity by placing humanity under its control. It needs an architecture in which protection remains bounded, context remains contestable and consequential authority remains owned by identifiable human beings.

The United States can contribute velocity and engineering capacity. China can contribute coordination and open technical alternatives. Europe can contribute rights, legal discipline and market-shaping standards. Britain can contribute security capability and stewardship. The Crown can hold the longer horizon of duty and continuity. The Vatican can illuminate the human purpose, dignity, subsidiarity and development that every technical architecture must serve.

Canada can connect these partial truths because it carries both moral inheritances, sits between American capability and European governance, and has already experienced the constitutional cost of emergency authority that moved faster than parliamentary and judicial correction. Canada does not lead by claiming moral purity. It leads by transforming hard institutional experience into a system that makes overreach more difficult and recovery more real.

The AI Stability Board can provide the platform. The reconstructed Zeroth Law can provide the mission. Dual Ethics can identify the two moral actors. Double-Bound AI Ethics can constrain power while developing human capability. Context Ownership, Decision Ownership, Mission Lineage and Mission Continuity can make the model operational.

The purpose of governance is not merely to constrain power. It is to direct power toward the improvement of humanity and the human person, and to cultivate people capable of exercising that power responsibly.

The greatest mistake in AI ethics is not that we failed to protect the human being. It is that, while protecting the human being, we stopped designing institutions that develop the human being.

Technological power does not confer the right to govern. Disarming AI is not enough. The world needs decision owners.

Sources and attribution

[1] CGTN, “Can China and the US reach consensus on AI guardrails?”, transcript reviewed September 22, 2026; discussion of nuclear non-proliferation, open-weight and closed models.

[2] CNBC and contemporaneous Dreamforce reporting on Jensen Huang, September 15-16, 2026.

[3] The White House, AI Action Plan; Executive Order 14409, June 2, 2026.

[4] European Union Artificial Intelligence Act and related implementation materials; European Commission risk-based AI governance framework.

[5] Transcript reviewed September 22, 2026, “AI became one of the main topics at the 81st UN General Assembly,” including British statements on standards, security and elected-government authority.

[6] The Royal Family, “The King convenes tech leaders for AI Summit in Scotland,” September 17, 2026; Court Circular, Forum on Artificial Intelligence and its role in serving the public good.

[7] Vatican, Pope Leo XIV, Magnifica Humanitas, May 15, 2026; public remarks by Chris Olah at its presentation.

[8] Center for Data Innovation, “AI Kill Switches Won’t Solve the Rogue AI Problem,” September 18, 2026.

[9] CNBC, “AI kill switch, explained: This simple safety solution may not work,” September 19, 2026.

[10] Historical sources on Catholic institutions and the Quiet Revolution in Quebec, including David Seljak, Michael Gauvreau and The Canadian Encyclopedia.

[11] Federal Court of Appeal, Canada (Attorney General) v. Canadian Civil Liberties Association, 2026 FCA 6, January 16, 2026.

[12] Government of Canada, Emergency Economic Measures Order, SOR/2022-22, registered February 15, 2022; Federal Court, Canadian Frontline Nurses et al. v. Canada (Attorney General), 2024 FC 42.

[13] Paul Samson, Centre for International Governance Innovation, “The World Needs an AI Stability Board,” September 18, 2026.

[14] Robert Fay, Centre for International Governance Innovation, “Digital Platforms Require a Global Governance Framework,” October 28, 2019; Financial Stability Board, institutional history.

[15] Government of Canada and European Commission materials on Canada-EU digital and AI cooperation, 2025-2026.

[16] Treasury Board of Canada Secretariat, Directive on Automated Decision-Making and Algorithmic Impact Assessment guidance.

[17] Isaac Asimov, Robots and Empire, Zeroth Law of Robotics.

[18] Supporting version history: Andre Rekhtine, The Biggest Mistake in AI Ethics, versions 1.5 and 2.3, September 2026.