From Safeguards to Moral Architecture

Last Updated May 9, 2026

Safeguards matter, but they are not enough. Institutions often respond to ethical risk by adding protective measures around systems that have already been designed, funded, authorized, and normalized: review steps, oversight committees, disclosure requirements, impact assessments, appeals channels, audit rights, red-team exercises, escalation procedures, or usage policies. These mechanisms are important. They can slow harm, detect failure, preserve evidence, and create some degree of accountability.

But safeguards are often treated as though they were the ethical core of responsible design, when in many cases they are only a boundary around choices whose deeper moral logic has already been fixed elsewhere. If the objectives of a system are poorly chosen, if its categories are unjust, if its incentives reward extractive or degrading conduct, or if its institutional setting is organized around opacity and asymmetrical power, then safeguards may reduce damage without altering the structure that produces it.

The ethical problem, in other words, does not begin only when a system malfunctions. It often begins with how the system has been imagined, ordered, authorized, measured, financed, and made durable.

To move from safeguards to moral architecture is to move from ethics as perimeter control to ethics as institutional design. It means asking not only what protections surround a system, but what kind of world the system helps make normal. It asks what purposes are built into the system, what forms of power it concentrates, what dependencies it creates, what harms it makes visible or invisible, what kinds of persons it assumes, and whether its ordinary operation is compatible with dignity, justice, accountability, ecological responsibility, and long-term public trust.

This article argues that ethical governance must move from safeguards to moral architecture. It examines the limits of safeguard thinking, clarifies what moral architecture means, explores why goals and objectives are already ethical choices, analyzes how institutional design distributes power, distinguishes compliance from moral formation, asks how systems recognize the dignity of those affected by them, shows why uncertainty requires humility rather than technocratic confidence, and explains why participation, public design, ecology, and long time horizons belong at the center of responsible governance.

Why This Belongs in Stewardship & Ethics

This article belongs in Stewardship & Ethics because it asks what responsible governance requires before harm occurs. Safeguards usually enter after a system has already been designed. Moral architecture asks whether the design itself is worthy of trust.

Stewardship is often associated with land, climate, biodiversity, and future generations. But stewardship also applies to institutions, technologies, administrative systems, public infrastructures, digital platforms, financial systems, and decision architectures. Any system that shapes human futures requires more than damage control. It requires ethical ordering.

The central question is not simply whether an institution has policies, review boards, audit logs, or appeals channels. The deeper question is whether the institution has been built around defensible purposes, accountable authority, meaningful participation, respect for dignity, humility under uncertainty, and responsibility across time.

A system can be heavily safeguarded and still be morally disordered. It can disclose risks while exploiting dependency. It can offer appeals while preserving unjust classifications. It can meet compliance requirements while eroding trust. It can monitor harm while continuing to optimize for goals that produce harm in ordinary operation.

Stewardship & Ethics therefore requires a deeper vocabulary than guardrails, controls, and mitigation. It requires attention to the moral structure of systems: the values they encode, the forms of life they normalize, the people they expose, the futures they make easier or harder, and the responsibilities they either clarify or obscure.

Back to top ↑

The Limits of Safeguard Thinking

Safeguard thinking often arises from a familiar institutional instinct. A new technology, administrative system, financial product, public program, or governance arrangement produces concern. In response, organizations add checks: compliance reviews, documentation requirements, fairness tests, escalation procedures, audit rights, red-team exercises, content policies, human review steps, incident reporting systems, or external advisory boards.

These interventions are not trivial. They can prevent some harms. They can make failures visible. They can create records for accountability. They can slow reckless deployment. They can give affected people some pathway for complaint or remedy. In many settings, stronger safeguards are urgently needed.

But safeguard thinking tends to assume that the underlying structure of the system is broadly legitimate and that ethics is mainly a matter of reducing excess, abuse, misuse, or operational error. This assumption is often too shallow.

A system can comply with formal safeguards while still organizing social life around degrading priorities. An institution can publish principles, run assessments, and document harms while continuing to optimize for extraction, manipulation, exclusion, or asymmetrical dependence. A platform can moderate abuse while preserving engagement-maximizing architectures that intensify addiction, outrage, surveillance, or social fragmentation. A public system can add appeals while preserving classifications that systematically misrecognize need. A financial institution can install disclosure regimes while maintaining incentives that reward short-term gain at the expense of social and ecological stability.

In such cases, safeguards do not fail because they are worthless. They fail because they are asked to compensate for a deeper disorder in purpose and design.

Safeguards can discipline execution, but they cannot by themselves justify the ends being pursued. They can identify harm, but they may not challenge the structure that produces it. They can create procedural legitimacy, but they may not restore moral legitimacy. They can reduce exposure, but they may not transform the underlying relation of power.

That is why ethical inquiry has to move beneath guardrails and ask whether the system deserves to exist in its present form at all.

Back to top ↑

What Moral Architecture Means

Moral architecture refers to the underlying normative structure of a system. It includes the goals a system is built to serve, the values those goals encode, the trade-offs made acceptable by design, the relations of power embedded in institutional procedure, and the assumptions about persons, risk, knowledge, nature, and responsibility that shape ordinary operation.

Architecture matters because it endures. It channels behavior before any single decision is made. It structures incentives, visibility, dependence, classification, authority, recourse, and accountability. It defines what can be recognized as a problem and what will remain outside the field of concern.

To speak of moral architecture is therefore to insist that ethics is not exhausted by prohibitions or constraints. Ethics is also constitutive. It enters at the level of design, mandate, purpose, institutional form, resource allocation, technical architecture, and decision procedure.

Moral architecture asks whether a system is ordered toward dignity, intelligibility, accountability, participation, proportionality, reciprocity, and care, or toward opacity, instrumentalization, dependency, asymmetry, and extraction. It asks not merely whether a given action is allowed, but what kinds of action the system rewards. It asks what burdens are shifted onto others, what risks are normalized, what forms of knowledge are privileged, and what conception of the human good is silently made operational.

Architecture is morally important precisely because it becomes ordinary. Once categories, incentives, and pathways of authority are built into routine operation, they cease to appear exceptional. They begin to define what the institution takes for granted.

Moral architecture therefore concerns the everyday ordering of institutional life, not only dramatic moments of failure.

Back to top ↑

Goals, Objectives, and the Ethics of Purpose

Every system is shaped by what it is designed to optimize. That fact alone makes ethics unavoidable. To choose an objective is to decide what counts as success. To choose a metric is to decide what aspects of reality will matter to the institution. To optimize for speed, engagement, prediction, cost reduction, throughput, revenue, risk scoring, enforcement efficiency, or behavioral conversion is not merely to define a technical problem. It is to rank goods, distribute burdens, and establish a hierarchy of values.

The moral question is therefore not simply whether a system achieves its stated objective efficiently. The question is whether the objective itself is worthy of governance.

This is one of the clearest points at which safeguards prove insufficient. If the governing objective is ethically distorted, safeguards may discipline execution while leaving purpose unchallenged. A system designed to intensify behavioral dependency may do so with extensive disclosure. A public tool designed around suspicion may process people consistently while still degrading them. A workplace system designed to maximize output at the cost of autonomy may include grievance procedures without addressing the deeper ordering of human life within the institution.

Ethical governance begins before compliance. It begins with the question of what the institution is trying to make normal.

Several questions should therefore be asked at the level of purpose:

• What does the system define as success?
• Who chose that definition?
• What goods are excluded from the objective function?
• What harms are treated as acceptable side effects?
• Who benefits when the system performs well?
• Who bears the costs of optimization?
• Can the system’s purpose be defended publicly to those affected by it?

Moral architecture demands that institutions submit objectives themselves to ethical scrutiny rather than assuming that control mechanisms can redeem any purpose so long as implementation appears careful enough.

Back to top ↑

Metrics, Categories, and Invisible Harm

Systems do not only act through decisions. They act through categories. They classify people, risks, needs, behaviors, assets, harms, outcomes, and priorities. These categories shape what can be seen, counted, governed, rewarded, punished, or ignored.

This is one of the most important dimensions of moral architecture. A system’s categories often determine the boundaries of institutional concern before any safeguard is applied. If a welfare system classifies people primarily as fraud risks, its architecture may produce suspicion even when procedural protections exist. If a workplace platform classifies workers primarily as units of productivity, it may degrade autonomy even if grievance mechanisms exist. If an environmental system classifies ecosystems only as service providers or carbon stores, it may miss cultural, spiritual, relational, and intergenerational meanings.

Metrics then reinforce categories. What is measured becomes governable. What is not measured can become invisible. Institutions that measure cost but not dignity, engagement but not manipulation, efficiency but not exclusion, output but not exhaustion, growth but not ecological depletion, and risk but not justice will often reproduce the harms they fail to see.

Safeguards may monitor the system’s outputs without questioning the categories that structure the outputs. Moral architecture goes deeper. It asks whether the system’s way of seeing the world is itself ethically adequate.

This matters because many harms are not exceptional incidents. They are ordinary consequences of narrowed perception. A system may work exactly as designed and still produce injustice because its categories are too crude, too suspicious, too extractive, too market-centered, too short-term, or too distant from lived experience.

A morally architected system must therefore make its classifications contestable. It must allow affected people to challenge not only individual outcomes, but the categories through which they are understood. It must ask what remains invisible when the system succeeds on its own terms.

Back to top ↑

Institutional Design and the Distribution of Power

No moral architecture can be understood apart from power. Systems distribute authority: who sets goals, who interprets rules, who sees what, who can challenge outcomes, who bears risk, who receives explanation, who has standing, and who remains exposed without meaningful remedy.

Architecture determines whether power is concentrated or shared, reviewable or insulated, transparent or obscure. It decides whether affected persons appear as participants, data points, users, claimants, risks, cases, customers, targets, or obstacles. In that sense, institutions do not simply exercise power. They design its pathways in advance.

This is why ethical governance requires more than adding external checks to a finished system. It requires attention to how authority is structured from the beginning.

Important questions include:

• Are there clear lines of responsibility, or has responsibility been diffused to the point of moral evasion?
• Are there meaningful channels for contestation, or only symbolic forms of review?
• Are people treated as subjects to whom justification is owed, or as objects to be processed efficiently?
• Are uncertainties acknowledged openly, or hidden beneath a rhetoric of precision?
• Can affected communities influence system design, or only respond after deployment?
• Can the system be paused, revised, or withdrawn when harm becomes clear?

Moral architecture takes these questions seriously because unjust power is often reproduced not through dramatic violations, but through ordinary institutional arrangements. A system may appear neutral because its authority has been embedded in procedure. But procedure itself can encode domination if the people affected by it lack voice, explanation, remedy, or power to challenge the terms of recognition.

Ethical disorder is frequently procedural and organizational before it is spectacular.

Back to top ↑

From Compliance to Formation

One of the dangers of safeguard language is that it encourages a compliance mentality. Ethics becomes something to satisfy, document, and pass, rather than something that forms institutional character.

The organization asks what controls are required, what disclosures are necessary, what audits must be completed, what policies will protect against reputational exposure, and what documentation will prove that responsible steps were taken. These are not unimportant questions. But when they dominate ethical reasoning, institutions begin to treat moral responsibility as a box to be checked rather than a discipline of judgment.

Moral architecture points in a different direction. It suggests that the ethical quality of an institution depends on what it is becoming through its routines, incentives, language, data practices, review mechanisms, and forms of accountability.

An institution shaped only by compliance may avoid the most visible failures while remaining indifferent to dignity, participation, social trust, ecological cost, or long-term harm. An institution shaped by moral architecture asks more demanding questions:

• What habits of decision-making are being cultivated?
• What types of conduct are being normalized?
• What dependencies are being entrenched?
• What responsibilities are being displaced?
• What kinds of persons are institutional actors becoming through repeated practice?

Ethics, in this sense, is formative before it is regulatory. It concerns the kind of institutional life being built, not only the kinds of incidents being prevented.

Compliance is necessary. But compliance without moral formation can produce institutions that obey the rules while losing the capacity for judgment.

Back to top ↑

Human Dignity and the Status of the Affected

A central test of moral architecture is how a system understands those who live under it. Are they conceived primarily as consumers, claimants, users, behavioral targets, risk profiles, cases, data subjects, or sources of friction? Or are they recognized as persons with standing, vulnerability, agency, and claims to explanation and recourse?

Safeguards often enter after this question has already been answered in practice. A system may provide channels of appeal while continuing to treat people as administratively burdensome. It may include consent interfaces while designing those interfaces to steer behavior. It may publish fairness commitments while leaving affected groups outside meaningful influence over design and review.

Moral architecture is more demanding because it asks whether the system has been ordered around a serious account of human dignity.

Dignity is not only a matter of avoiding overt abuse. It also concerns whether people are treated as beings to whom reasons are owed, whose vulnerability matters, whose participation has standing, and whose lives are not reducible to the efficiency needs of institutions.

A morally serious architecture will therefore design for:

• intelligibility: people can understand how decisions affecting them are made;
• contestability: people can challenge decisions, classifications, and assumptions;
• proportionality: institutional power does not exceed what is necessary and justified;
• recourse: people have meaningful pathways to remedy;
• respect: systems do not degrade people through suspicion, manipulation, opacity, or unnecessary dependency;
• participation: affected communities help shape the systems that shape them.

This is the difference between designing around harm prevention alone and designing around moral status. The former asks how to limit damage. The latter asks what kind of regard the institution owes to the people whose lives it helps structure.

Back to top ↑

Uncertainty, Risk, and the Discipline of Humility

Another difference between safeguards and moral architecture lies in their treatment of uncertainty. Safeguards often presume that risks can be identified, catalogued, measured, and managed within existing institutional assumptions. Moral architecture begins from a more disciplined insight: systems often operate under conditions of uncertainty that cannot be fully tamed by measurement or prediction.

Models can be wrong. Environments can shift. Categories can misfire. Incentives can change behavior. Interactions between technical and social systems can produce novel forms of harm. A system that works in one setting may fail in another. A tool that appears fair under one metric may reproduce injustice under another. A governance mechanism that looks adequate at small scale may become dangerous when generalized.

This is not an argument for paralysis. It is an argument for humility in design.

Humility has architectural implications. It requires documentation that makes assumptions visible rather than burying them. It requires review mechanisms that remain open to revision rather than presenting systems as closed achievements. It requires escalation pathways, independent scrutiny, and the capacity to pause, amend, or withdraw systems whose harms prove more serious than expected.

Above all, humility requires institutions not to confuse operational competence with moral certainty.

A system that cannot admit uncertainty is not robust. It is brittle in a morally dangerous way. Architecture becomes more trustworthy not when it claims perfect foresight, but when it creates disciplined ways of learning, revising, correcting, and being held accountable under conditions that remain partly unknowable.

Back to top ↑

Participation and the Moral Order of Design

Architectural ethics also changes the role of participation. Under a safeguard model, participation is often invited late and narrowly. Stakeholders are asked to comment on a system whose aims, categories, and institutional commitments are already largely fixed. Their role is advisory at the margins.

Under a moral-architectural model, participation enters earlier and more deeply. Affected persons and communities help shape the understanding of the problem, the choice of objectives, the identification of risks, the meaning of fairness, and the conditions under which review should occur.

This matters because exclusion is not only a procedural defect. It is often a sign that the architecture itself has been built around the perspective of those already closest to power.

When that happens, harms may appear only after deployment because they were invisible to the designers’ moral horizon from the beginning. Participation cannot solve every structural problem, but it is one of the few ways institutions can resist becoming closed moral worlds in which their own assumptions appear universal.

A morally serious architecture therefore does not treat participation as a legitimizing ritual. It treats participation as part of the epistemic and ethical substance of responsible design.

Participation belongs at the center because systems that govern people should not be designed as though only the already powerful are capable of understanding what justice requires.

Back to top ↑

Public Institutions and the Ethics of Constitutional Design

In public institutions, the shift from safeguards to moral architecture is especially important. State systems do not merely offer products or services. They exercise authority over rights, status, recognition, access, obligation, punishment, benefits, eligibility, movement, and protection.

For that reason, the ethics of public design cannot be satisfied by adding oversight onto administratively convenient systems. Public architecture must be ordered toward legality, due process, equal standing, reason-giving, reviewability, proportionality, and public justification from the beginning. Where these conditions are absent, safeguards may soften the edges of domination without altering its basic structure.

This is why digital government, administrative systems, automated public decision-making, risk scoring, eligibility systems, and public-sector AI raise constitutional as well as managerial questions. A system that is efficient but opaque, consistent but unreviewable, or scalable but degrading is not redeemed by the existence of a complaints form alone.

Public moral architecture must ask what kind of state is being built through its ordinary technical systems. Does it enhance civic standing and trust, or expand distance, suspicion, and asymmetrical dependence? Does it help people understand and contest public authority, or does it make authority more difficult to see? Does it improve access, or does it create a new layer of administrative exclusion?

The question is not whether public institutions should use complex systems. The question is whether those systems are worthy of democratic authority.

Governance becomes ethically serious when technical design is treated as part of constitutional order rather than as a merely operational matter.

Back to top ↑

AI, Automation, and Emerging Systems

The language of safeguards is especially common in discussions of artificial intelligence and emerging technology. Developers and institutions often emphasize risk management frameworks, model cards, audits, impact assessments, bias testing, red-teaming, human oversight, disclosure rules, content policies, and monitoring systems. These tools are valuable. Responsible AI systems need them.

But AI governance also reveals the limits of safeguard thinking. An AI system can be monitored, documented, benchmarked, and constrained while still being built around a questionable purpose. A system designed to automate surveillance, manipulate behavior, intensify workplace control, or classify vulnerable people through crude proxies may remain morally defective even if it performs within technical risk parameters.

The architectural question comes first: should this system exist, in this institutional setting, for this purpose, under this authority, with these incentives, affecting these people, under these conditions of contestability and remedy?

AI makes this question urgent because automated systems can scale institutional assumptions quickly. They can make categories more durable, decisions more opaque, authority more distant, and accountability more difficult to locate. They can also improve public services, accessibility, scientific discovery, environmental monitoring, and administrative efficiency when designed with care, restraint, and accountability.

The ethical difference lies in architecture.

A morally serious approach to AI and emerging systems should ask:

• Is the objective publicly defensible?
• Are affected people treated as participants or merely as data subjects?
• Can decisions be explained, challenged, corrected, and reversed?
• Are uncertainty and error honestly represented?
• Does the system reduce burdens or shift them onto the vulnerable?
• Does automation clarify responsibility or obscure it?
• Can the system be paused or withdrawn when harms become clear?

AI governance cannot be reduced to safer deployment of any system an institution wishes to build. It must include judgment about purpose, legitimacy, power, dignity, and the futures being normalized through automation.

Back to top ↑

Ecology, Time, and the Wider Field of Responsibility

Safeguards are often short-horizon instruments. They focus on immediate risk, near-term compliance, and recognizable incidents of harm. Moral architecture widens the temporal field. It asks how systems shape long-term dependence, institutional habits, ecological burden, social trust, and intergenerational consequence.

A design that appears acceptable under a narrow risk lens may still be ethically disordered if it externalizes ecological cost, entrenches surveillance, normalizes administrative suspicion, degrades labor, intensifies resource extraction, or locks vulnerable populations into brittle infrastructures with little voice over future change.

This longer view is essential because many of the most serious ethical failures are cumulative rather than dramatic. They emerge through normalization:

• normalization of opacity;
• normalization of extractive data relations;
• normalization of administrative suspicion;
• normalization of degraded labor;
• normalization of ecological externalization;
• normalization of short-term optimization;
• normalization of dependence without recourse.

Moral architecture therefore requires stewardship across time. It asks not only whether a system can be controlled today, but whether the world it is helping to build remains one in which dignity, accountability, ecological continuity, and democratic agency are still possible tomorrow.

Ethical governance becomes more demanding at this point because it must consider not only immediate harm prevention but the habits, infrastructures, and dependencies that systems sediment over time.

Architecture matters because it makes futures easier or harder to reverse.

Back to top ↑

Toward Morally Architected Systems

To move from safeguards to moral architecture is not to abandon controls, audits, or oversight. It is to relocate them within a deeper ethical order.

Safeguards remain necessary. But they must be attached to institutions and systems whose purposes, power structures, and accountability relations have themselves been subjected to moral scrutiny.

A morally architected system should include several features:

• Defensible purpose: the system’s objective should be ethically justifiable, not merely efficient or profitable.
• Accountable authority: responsibility should be clear, reviewable, and impossible to hide behind technical complexity.
• Participatory design: affected communities should shape problem definition, objectives, risk identification, and review.
• Dignity-centered operation: people should be treated as persons owed reasons, not merely as users, cases, risks, or data points.
• Contestability and recourse: decisions, classifications, and assumptions should be challengeable and correctable.
• Humility under uncertainty: systems should be designed for revision, pause, withdrawal, and learning.
• Long-horizon responsibility: ecological, institutional, and intergenerational effects should be part of design evaluation.
• Public intelligibility: affected people should be able to understand the system well enough to hold it accountable.

Such a shift is demanding because it requires institutions to think ethically before failure rather than merely after it. It requires them to ask whether a system deserves to exist in its present form, not only whether it can be rendered acceptable through controls.

It also requires a more serious understanding of governance itself. Governance is not only the management of risk. It is the shaping of institutions so that power operates in ways compatible with dignity, justice, intelligibility, accountability, and stewardship.

A system with excellent safeguards but poor moral architecture may be less dangerous than one with neither, but it is still ethically unfinished. The task is not to choose between guardrails and design. The task is to make design itself answerable to a more serious moral order.

Back to top ↑

Moral Architecture Diagnostic Table

Back to top ↑

Conclusion: From Safeguards to Moral Architecture

The language of safeguards has been useful because it recognizes that systems can do harm and that institutions need mechanisms to detect, prevent, and correct failure. Safeguards remain necessary. Institutions need audits, reviews, appeals, documentation, monitoring, red-teaming, incident response, and oversight.

But safeguards are only part of the ethical task. They address conduct within a structure. They do not by themselves justify the structure.

The deeper question is architectural: what values are built into the system, how authority is distributed, what forms of standing are recognized, what harms are made normal, what futures are being locked in, and whether the institution is ordered toward something worthy of human trust.

To move from safeguards to moral architecture is therefore to move from ethics as perimeter control to ethics as institutional design. It is to recognize that responsibility is not fulfilled by surrounding power with rules while leaving its deeper purposes unexamined. Moral seriousness begins earlier. It begins in the ordering of objectives, the design of accountability, the treatment of persons, the admission of uncertainty, the protection of ecological and social futures, and the willingness to build systems whose ordinary operation is aligned not merely with efficiency or compliance, but with dignity, justice, and stewardship.

The ethical future of institutions will not be determined only by how many guardrails they add. It will be determined by what those institutions are built to serve, how they distribute power, whether they can be challenged, whether they treat people as persons rather than inputs, and whether they help create a world in which responsibility is built into the architecture rather than appended after the fact.

Safeguards matter. But moral architecture determines what safeguards are trying to protect.

Back to top ↑

Related Reading

• Institutional Stewardship, Governance, and Public Trust
• Authority and Legitimacy in Institutions
• Participation, Accountability, and Procedural Justice
• Ethics of System Design
• Risk Governance in Complex Societies

Back to top ↑

Further Reading

• National Institute of Standards and Technology (NIST) (2023) Artificial Intelligence Risk Management Framework (AI RMF 1.0). Gaithersburg, MD: NIST. Available at: https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf
• Organisation for Economic Co-operation and Development (OECD) (2024) Framework for the Anticipatory Governance of Emerging Technologies. Paris: OECD. Available at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/framework-for-anticipatory-governance-of-emerging-technologies_14bf0402/0248ead5-en.pdf
• Organisation for Economic Co-operation and Development (OECD) (2025) Governing with Artificial Intelligence. Paris: OECD. Available at: https://www.oecd.org/en/publications/governing-with-artificial-intelligence_26324bc2-en.html
• Organisation for Economic Co-operation and Development (OECD) (n.d.) Technology Governance. Available at: https://www.oecd.org/en/topics/technology-governance.html
• UNESCO (2021) Recommendation on the Ethics of Artificial Intelligence. Paris: UNESCO. Available at: https://unesdoc.unesco.org/ark:/48223/pf0000380455
• UNESCO (n.d.) Ethics of Artificial Intelligence. Available at: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
• Floridi, L. and Cowls, J. (2019) ‘A unified framework of five principles for AI in society’, Harvard Data Science Review, 1(1). Available at: https://hdsr.mitpress.mit.edu/pub/l0jsh9d1/release/8
• Winner, L. (1980) ‘Do artifacts have politics?’, Daedalus, 109(1), pp. 121–136. Available at: https://www.jstor.org/stable/20024652

Back to top ↑

References

• Floridi, L. and Cowls, J. (2019) ‘A unified framework of five principles for AI in society’, Harvard Data Science Review, 1(1). Available at: https://hdsr.mitpress.mit.edu/pub/l0jsh9d1/release/8
• National Institute of Standards and Technology (NIST) (2023) Artificial Intelligence Risk Management Framework (AI RMF 1.0). Gaithersburg, MD: NIST. Available at: https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf
• Organisation for Economic Co-operation and Development (OECD) (2024) Framework for the Anticipatory Governance of Emerging Technologies. Paris: OECD. Available at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/framework-for-anticipatory-governance-of-emerging-technologies_14bf0402/0248ead5-en.pdf
• Organisation for Economic Co-operation and Development (OECD) (2025) Governing with Artificial Intelligence. Paris: OECD. Available at: https://www.oecd.org/en/publications/governing-with-artificial-intelligence_26324bc2-en.html
• Organisation for Economic Co-operation and Development (OECD) (n.d.) Technology Governance. Available at: https://www.oecd.org/en/topics/technology-governance.html
• UNESCO (2021) Recommendation on the Ethics of Artificial Intelligence. Paris: UNESCO. Available at: https://unesdoc.unesco.org/ark:/48223/pf0000380455
• UNESCO (n.d.) Ethics of Artificial Intelligence. Available at: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
• Winner, L. (1980) ‘Do artifacts have politics?’, Daedalus, 109(1), pp. 121–136. Available at: https://www.jstor.org/stable/20024652

Back to top ↑