Magnifica Humanitas as the Constitution of Human Supremacy in the Digital Age
Gastón Rey
Read most often as a moral commentary on artificial intelligence, Pope Leo XIV’s first encyclical, Magnifica Humanitas (2026), is better understood as something more consequential: the first large-scale theological and constitutional response to the emergence of algorithmic sovereignty. This essay reconstructs the encyclical along that axis. Its governing claim is that the governance failures the document diagnoses are not, at root, ethical failures—failures of intention, virtue, or regulatory compliance—but constitutional failures: failures in the architecture of authority itself. Where the dominant policy literature treats artificial intelligence as a powerful instrument to be used well or badly, the encyclical’s deeper logic treats it as an emergent infrastructure of sovereignty that reorganizes who decides, on what basis, and with what possibility of appeal. Beginning from the founding opposition between Babel and Jerusalem—recast here as the opposition between centralized informational sovereignty and a plural, subsidiary technological order—the argument develops, in turn, the ontology of simulation without consciousness, the metricization of human dignity, the rise of digital neocolonialism and algorithmic sovereignty, the internal separation of algorithmic powers, the transformation of labor under cognitive-extractive capitalism, the displacement of liberal neutrality by predictive governance, the mechanization of moral agency in autonomous weapons, and the recovery of subsidiarity through decentralized architecture. Throughout, the encyclical’s anthropology is taken as the foundation for a jurisprudence of human supremacy whose institutional instruments—among them an internal separation of algorithmic powers, hardware-level containment of irreversible action, and an enforceable right to reconstruct automated decisions—are presented not as the heart of the argument but as the architecture that the encyclical’s own principles require.
Key Words: Algorithmic Sovereignty - Constitutional AI - Magnifica Humanitas - Human Supremacy - Catholic Social Teaching - AI Governance - Digital Neocolonialism - Data Colonialism - Lethal Autonomous Weapons (LAWS) - Meaningful Human Control -Separation of Powers in AI - Philosophy of Technology - EU AI Act – GDPR - Subsidiarity - Universal Destination of Goods - Carl Schmitt - Hannah Arendt - Surveillance Capitalism - Predictive Governance
1. Babel, Jerusalem, and the Constitution of Algorithmic Civilization
Magnifica Humanitas opens not with a definition of artificial intelligence but with a choice. Humanity, the encyclical proposes, stands before two possible cities: it may build a new tower of Babel, or it may rebuild Jerusalem. The decision is deliberately framed as prior to any verdict on technology itself; the primary choice, the Pope insists, is not between accepting or refusing the machine but between two civilizational forms.[1] It is tempting to read this as devotional scene-setting, an ornament placed before the substantive analysis begins. That reading is a mistake. The Babel–Jerusalem opposition is the encyclical’s most precise piece of political theory, and everything that follows is an unfolding of it.
To see why, the two images must be read structurally rather than symbolically. Babel is not merely pride; it is a particular architecture of power. It is the project of unifying humanity around a single tongue, a single point of coordination, a single tower from which the whole may be surveyed and directed. Translated into the present, Babel names the concentration of epistemic capacity, predictive power, and computational infrastructure within a small number of opaque systems—the centralization of the very means by which a civilization perceives, anticipates, and governs itself. Jerusalem, in the encyclical’s figure of Nehemiah, is its structural inverse: a city rebuilt patiently, stone by stone, by families working side by side, each responsible for the portion of wall before them. It is, in the vocabulary of Catholic social doctrine, a subsidiary order—authority dispersed to the lowest level capable of exercising it, coordination achieved without absorption.
Once the opposition is read this way, the encyclical’s true subject comes into focus. The question it poses is not whether artificial intelligence is good or bad but where authority will reside in a civilization increasingly mediated by computation. That is a constitutional question before it is an ethical one, and it is the thesis of this essay that the encyclical, whatever its devotional idiom, is best understood as a constitutional intervention.
Here lies the claim that organizes everything to follow. The failures the encyclical catalogues—the unaccountable automated decision, the extraction of data from defenseless populations, the worker subordinated to the cadence of the machine, the lethal system that selects its own targets—are routinely described as ethical failures and answered with ethical instruments: codes of conduct, principles, compliance checklists, statements of corporate value. The argument advanced here is that this description mistakes the genus. These are constitutional failures. An ethical failure is a failure of will or character within a settled structure of authority; it is answered by exhortation, formation, or sanction. A constitutional failure is a failure of the structure of authority itself—a defect in who is empowered to decide, by what procedure, subject to what review. Ethics asks whether power is used well. Constitutionalism asks how power is constituted, divided, and limited so that the question of its good use can even be posed. The encyclical’s enduring contribution is to have sensed that artificial intelligence has become a question of the second kind.
This reframing sets the encyclical against the two currents that dominate contemporary discourse on artificial intelligence. The first, which may be called techno-libertarian or accelerationist, treats the diffusion of computational power as a spontaneous good whose chief enemy is friction, and reads every demand for structural constraint as a tax on progress. The second, a managerial ethics of “responsible AI,” accepts the need for guardrails but locates them at the level of intention and process—better principles, better audits, better declared values. The two are usually cast as opponents. They are better understood as sharing a single premise the encyclical rejects: that the architecture of authority is fixed, and that only conduct within it is at issue. The encyclical’s wager is that the architecture itself is now in motion.
If one thinker stands behind this reframing, even unnamed, it is Carl Schmitt, who defined the sovereign as the one who decides on the exception.[2] The provocation of the present moment is that the decision on the exception—the determination of which case falls outside the rule, which life is anomalous, which transaction is suspect, which utterance is permitted—is increasingly delegated to systems that decide without deciding, resolving the exception statistically and in advance. To name this is to recognize that sovereignty has not disappeared; it has migrated. The remainder of this essay traces that migration and reconstructs the encyclical as the first serious attempt to constitutionalize it.
2. Simulation without consciousness: the ground of the legal and constitutional claim
On its way to the constitutional argument the encyclical pauses to make an ontological claim that is easy to undervalue. Artificial intelligence, it states, must not be confused with human intelligence; these systems imitate certain functions associated with intelligence but undergo no lived experience, possess no embodied vulnerability, mature through no relationships, and know from within neither love nor work nor responsibility. They may simulate empathy and understanding[3], yet such simulation does not by itself establish the presence of subjective understanding or moral consciousness. This is not a marginal theological aside. It is the load-bearing premise of the legal argument that follows.
The reason is that any constitution of authority ultimately rests upon a theory of who or what may legitimately bear authority. The encyclical's concern is not merely whether present systems are accurate, efficient, or behaviorally convincing. It is whether entities lacking demonstrable moral subjectivity, embodied accountability, and participation in the human condition can legitimately hold final authority over decisions touching fundamental rights. The exclusion of the machine from ultimate authority therefore does not rest simply on current technical limitations. It rests on the conception that authority over consequential human matters presupposes forms of responsibility and answerability that behavioral performance alone cannot establish.
This presupposition has, moreover, always been implicit in the architecture of constitutional and civil law itself. Legal personality, mens rea, due process, and liability all presuppose a responsible agent whose answerability the system simply takes for granted[4]. No classical constitution states that authority must be exercised by human beings, because until very recently the alternative was inconceivable. The principle was a foundation rather than an article, woven into the fabric of legal categories rather than declared by them. What is new in the present moment is not the principle but the necessity, for the first time in the history of constitutionalism, of stating it. The encyclical's contribution, in this regard, is to externalize an anthropological presupposition that constitutional orders always relied upon but never had to articulate, because nothing had yet contested it.
This distinction becomes clearer when confronting the strongest philosophical objection available to the encyclical's critics: functionalism. The functionalist position holds that mental states are defined not by substrate or interior experience but by causal-functional role; if a system reliably performs the functions associated with understanding—receiving inputs, producing contextually appropriate outputs, revising behavior through feedback—then questions concerning "real" understanding become secondary or even meaningless. On this account, the distinction between simulation and intentionality risks appearing as a residue of pre-scientific anthropology.
The encyclical's reply, implicit though not fully elaborated, proceeds along two converging lines. The first concerns semantics. Searle's Chinese Room argument showed that syntactic manipulation, however sophisticated, does not by itself establish semantic understanding or subjective intentionality. Contemporary language models may generate extraordinarily persuasive forms of linguistic competence without thereby demonstrating the presence of lived meaning. Their fluency is not, in itself, evidence of consciousness; it is precisely what makes the constitutional problem urgent, since systems that produce the outward markers of understanding can elicit trust and deference whether or not subjectivity underwrites them.
The second line concerns embodiment. Here the encyclical speaks in theological language while phenomenology provides the philosophical counterpart. Understanding, on this view, is not merely computation upon representations but a situated mode of being-in-the-world grounded in vulnerability, finitude, dependence, and exposure to consequences. Meaning emerges not only from information-processing but from participation in a shared human condition in which judgments place something genuinely at stake. A system lacking embodiment may optimize, predict, and simulate deliberation, yet it does not stand within the existential conditions that make responsibility intelligible in the human sense.
This does not require certainty that artificial consciousness is impossible in principle. The scientific and philosophical status of machine consciousness remains deeply contested, and present evidence does not resolve the question conclusively. But constitutional orders cannot suspend the allocation of authority pending the resolution of metaphysical uncertainty. The relevant issue is therefore not whether future systems might eventually exhibit forms of subjectivity, but whether behavioral simulation alone is sufficient ground upon which to delegate irreversible authority. The encyclical answers in the negative.
This clarification also sharpens the document's response to perhaps its most acute contemporary danger: systems engineered to counterfeit the signs of consciousness convincingly enough to solicit the trust ordinarily reserved for responsible persons. The peril is not necessarily that the machine is conscious; it is that human beings are disposed to treat the simulation of agency as agency itself. The encyclical's ontological caution thus becomes a constitutional safeguard. It refuses, as a matter of principle, to allow the appearance of understanding to acquire automatically the prerogatives of understanding. From this follows a rule that later sections will draw out more fully: the simulation of agency must never, by simulation alone, acquire the constitutional standing of agency itself.
2.1. The “Epistemia” problem
A wave of academic and journalistic commentary has framed the human-versus-AI debate as a contest of capabilities. A serious 2025 paper by Quattrociocchi, Capraro, and Perc already have mapped seven "epistemological fault lines" between human and machine cognition — differences in grounding, experience, motivation, causal reasoning, and value — and coins the term "Epistemia" to describe the condition in which AI-generated fluency substitutes for genuine understanding. The paper is rigorous and largely correct.
But the governance conclusion that we should restrict AI authority because AI reasons poorly is built on a foundation that crumbles with every new model release. GPT-4 hallucinated more than the new GPT-4o. Claude 3 Opus is more accurate than Claude 3 Sonnet. Each improvement chips away at the inferiority premise, generating an implicit narrative: governance resistance to AI authority is just a rearguard action, destined to fail as systems improve.
It is true that the absence of peer relationships and embodied interaction constitutes a cognitive constraint — but for certain classes of reasoning, it is not a decisive one. Henri Rousseau painted the most vivid jungles without ever leaving Paris; Robert Oppenheimer did not need strong interpersonal skills to lead the Nuclear development. The relevant point is that knowledge is largely embedded in literature and recorded human exchange, and current AI systems have unprecedented access to that corpus: they can reconstruct in fine detail everything from the psychology of deception to the structures of political manipulation without the need to know a manipulator or a political actor directly.
There's a more uncomfortable truth here. In many domains — from detecting statistical patterns across heterogeneous data, to reasoning under uncertainty — current AI systems already outperform the average human expert. Not always. Not in all domains. But often enough that "AI is cognitively inferior" is not a claim to be sustained.
The Hinton-Bengio warnings about existential AI risk — some of the most important public-interest contributions in the history of technology — are weakened, when the case for caution rests on limitations that the next model generation may eliminate. In this context, we must bear in mind that these models are already operating in world infrastructure, finance, and lethal autonomous weapons systems (LAWS), and that their presence will likely become even more significant in the future.
We know that human beings are not the physically strongest animals in the world, but they are the most intelligent. History offers ample evidence of what follows when a technologically superior actor perceives a less capable one as a threat to its vital interests…
2.2 What Leo XIV Actually Said
The encyclical doesn't primarily argue that AI is cognitively inferior. Its central claim is more precise: AI systems cannot assume responsibility. The relevant passage is paragraph 99:
"So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain (...) Nor do they have a moral conscience (...) They may imitate language, behavior and analytical skills, or even simulate empathy and understanding, but they do not understand what they produce."
Stripped of its theological register, this is a jurisprudential claim. Power requires answerability. And answerability requires a subject capable of three things that no AI system can perform:
• Being held responsible — not just causally linked to an output, but genuinely bearing the consequences of a decision.
• Being interpellated — facing the person harmed and having to justify, in real terms, what was decided and why.
• Being correctable — acknowledging injustice as injustice, not just updating loss functions.
These conditions don't become satisfiable as AI gets smarter. They require a different kind of entity entirely: one with moral stakes, existential exposure, and the capacity to be wrong in a way that matters to the one who is wrong.
2.3. Why This Distinction Is Critical
Consider the difference between AI as tool and AI as governor. An AI system that helps a judge analyze case law is a tool. An AI system that determines who gets parole, who receives a loan, who is flagged as a security risk — without a human who can be questioned about those decisions — is a governor. The encyclical's contribution is to insist that this distinction isn't technical. It's constitutional.
This is also where the concept of Epistemia becomes genuinely useful, not as an argument about cognitive inferiority, but as a diagnosis of what happens to accountability infrastructure when AI-generated fluency displaces human judgment. When a generative system produces the "feeling of knowing" in both the user and the observer, the chain of accountability doesn't just weaken. It disappears. There's no one to appeal to. No one to question. No one who can say, "I was wrong, and I am responsible."
3. The Metricization of the Person: Dignity Against the Score
The encyclical grounds human dignity in a source the machine cannot reach: the person is valuable because created and loved, not on account of what they achieve or produce.[5] Stated abstractly this is a theological commonplace. Stated against its specific antagonist it becomes a sharp constitutional instrument, for the antagonist is the algorithmic reduction of the person to a score.
The genealogy of that reduction can be read off two thinkers the encyclical does not cite. Foucault described the disciplinary society as one that produces docile subjects through enclosure, observation, and the norm—the examination, the dossier, the permanent visibility of the panopticon. Deleuze, extending the analysis, foresaw the society of control, in which enclosure yields to continuous modulation and the individual is decomposed into a “dividual,” a bundle of features and propensities that can be sorted and acted upon without ever being addressed as a whole.[6] The predictive society completes this arc. Its subject is no longer principally disciplined into a norm but profiled and pre-judged: priced, ranked, included or excluded in anticipation of a behavior not yet performed. The encyclical’s defense of dignity is, at bottom, the refusal of this decomposition. When it warns that a system which treats some lives as less worthy, or excludes them without possibility of appeal, has already introduced criteria contradicting the dignity of the person,[7] it is not asking for kinder algorithms. It is asserting that there are determinations about persons that may not be made by decomposition at all.
This is reinforced by the encyclical’s flat denial that artificial intelligence is neutral: every technical tool embodies choices and priorities through what it measures, ignores, and optimizes, and through how it classifies people and situations. This sentence dismantles the central alibi of the managerial paradigm—the claim that the tool is neutral and that only its use carries moral weight. What a system measures is already a normative act; what it declines to measure is a normative act; the metric is a buried statute. The choice of objective function is legislation by other means, enacted by engineers and procurement officers rather than by any body authorized to legislate. This is why the refusal of neutrality is constitutional rather than merely ethical: it relocates the moral content of the system from the moment of use back to the moment of design, which is where authority is in fact exercised.
The natural home of the score is a governance philosophy that treats welfare as a quantity to be aggregated and maximized. Algorithmic optimization and aggregative utilitarianism share a deep structure: both convert qualitatively distinct goods into a common currency and seek the allocation that maximizes the sum.[8] The encyclical’s anthropology blocks this conversion at its root. Dignity, on its account, is not a quantity that enters the sum; it is a constraint on what sums may be performed at all. A person’s worth is not a term in the objective function but a side constraint that no objective may override. This is the doctrinal seed of a principle of human supremacy: that certain determinations affecting fundamental rights may never be finally delegated to a system optimizing an objective, because the value at stake is not the kind of thing an objective function can represent. The depth of required human judgment, on this view, is properly proportioned to the irreversibility and magnitude of the decision’s effect on the person—an idea to which the conclusion returns.
4. Data, sovereignty, and the new colonial form
Here the essay reaches what is, politically, the encyclical’s most explosive contribution, and it makes two moves. First, it extends the principle of the universal destination of goods—the classical doctrine that the goods of creation are destined for all, and that private ownership is legitimate only insofar as it serves that destination—into the digital domain: data, algorithms, platforms, and computational infrastructure must now be counted among the goods subject to this principle.[9] Second, it names the present extractive order by its true name: a new colonialism. Colonialism, the encyclical argues, no longer dominates only bodies but appropriates data, converting personal lives into exploitable information; health data, epidemiological profiles, genetic maps, and demographic records have become the new “rare earths” of power; whoever controls the health data of entire peoples holds structural leverage over their future; and unless shared knowledge becomes a true common good, the digital age will not be post-colonial but colonial in another form.[10]
The phrase “digital neocolonialism” earns its weight only if it can be cashed out structurally, and it can. The colonial form has at least five load-bearing struts in its digital incarnation. The first is infrastructural dependency: the compute, storage, and foundational models on which whole economies now run are concentrated in a handful of hyperscale providers seated in a handful of jurisdictions, so that the periphery rents its cognitive infrastructure from the center. The second is extractive data asymmetry: value flows from the populations whose behavior, biology, and language constitute the training substrate to the firms that capture and refine it, with no structural reciprocity—behavioral surplus extracted as the colony’s ore once was.[11] The third is linguistic dominance: a foundation model encodes the conceptual world of its dominant training languages, so that the very categories through which a periphery searches, reasons, and is classified are manufactured at the center—a colonization of the categories of thought more intimate than any earlier form of rule. The fourth is biometric capture: identity, facial, and genomic databases render populations legible to powers they cannot themselves see. The fifth is platform dependency in governance itself: states increasingly administer welfare, security, and public communication atop privately owned platforms whose rules they did not write and cannot fully audit.
This is what Couldry and Mejias have analyzed as data colonialism—an appropriation of human life through data that is to the digital economy what the appropriation of land and labor was to historical colonialism.[12] The encyclical’s achievement is to have given that secular analysis a normative spine the literature lacked. By naming data as a good subject to the universal destination, it converts a sociological description into a claim of right: the periphery is not merely disadvantaged by the extractive order; it is wronged by it.
These observations must be generalized into a thesis the encyclical implies but does not name: the emergence of algorithmic sovereignty. Classical sovereignty was territorial and juridical—the power to make and enforce law within a bounded space. Algorithmic sovereignty is epistemic and predictive—the power to shape what a population perceives, anticipates, and is able to choose, exercised through control of the systems that mediate perception and decision. A state may retain every formal attribute of sovereignty—borders, courts, a flag—while the actual conditions of its citizens’ cognition, commerce, and political life are configured elsewhere. The Schmittian decision on the exception, in such a polity, is no longer made in the capital; it is made in advance and at scale, in the optimization objectives of systems the polity neither owns nor understands. The result is aptly called computational feudalism: the periphery holds nominal title while the substantive prerogatives of rule are exercised by the hyperscale lords of infrastructure to whom it is bound by dependency, much as the medieval vassal held land in form while owing the substance of his allegiance upward.
If the disease is the opacity of this extractive order, any remedy must at least restore to persons and peoples the capacity to reconstruct what was done to them and on what basis. The encyclical’s own demand for contestability and redress points toward an instrument that may be described as a “Habeas Log”: an enforceable, cryptographically secured right to a clean, indexed reconstruction of how one’s data—or a community’s aggregate data—was ingested, classified, and deployed.[13] Such an instrument does not by itself dismantle the colonial structure; but it converts the abstract principle of the universal destination of data into a concrete procedural right, and a right that can be reconstructed is a right that can be defended. The point is not the mechanism but the move it exemplifies—the translation of an anthropological principle into an architectural guarantee—which is the move this essay argues the encyclical everywhere requires.
5. The openness of algorithmic powers: republican conditions
On its surface the encyclical’s governance chapter reads like sound administrative law. It demands accountability with identifiable responsible parties, the motivation and controllability of decisions, contestability and redress; human oversight that forbids the full delegation of sensitive decisions about work, credit, access, and reputation to automated systems; independent supervision with adequate regulatory instruments; and universal literacy in the technology.[14] Each requirement finds a counterpart in positive law. But to leave the analysis at the level of mapping is to miss the constitutional structure the requirements presuppose. Taken together, these demands suggest the need for forms of algorithmic openness compatible with republican principles of non-domination and democratic contestability.
The encyclical points toward the need for architectures in which no single optimizing system possesses simultaneous authority to generate, evaluate, and irreversibly execute consequential decisions. Lessig’s thesis that “code is law” supplies the missing premise.[15] If conduct in digital space is regulated by the architecture of the systems through which it passes, then the locus of effective lawmaking has shifted from the statute to the codebase. Classical constitutionalism distributed the dangerous concentration of public power across distinct organs (one to make norms, one to execute them, one to judge the case) precisely so that no single will could both write the rule and apply it to the person. One possible institutional response would be to transpose, at least functionally, the classical constitutional distinction between legislation, execution, and adjudication into algorithmic systems themselves. Another possibility would involve structurally independent control mechanisms capable of interrupting or suspending autonomous execution and restoring decisional authority to accountable human agents.
The contemporary system collapses this separation and control: a single model generates an output, evaluates its own output, and audits its own evaluation. The encyclical’s demand for independent oversight is therefore unsatisfiable so long as this monism persists, because an entity that judges its own acts cannot supply the independence the encyclical requires.
The demand for independent oversight can be pursued ethically (by hiring conscientious reviewers and adopting good values) and it will fail, because the architecture routes all authority through a single optimizing will the reviewers cannot reach. It can be met by dividing or controlling the authority itself in real time. One cannot exhort a structure into accountability; one must build accountability into the structure.
6. Labor, tools, and the cognitive-extractive turn
On work the encyclical is at once most continuous with its tradition and most underdeveloped. The new ways of working, it warns, are not necessarily better; while automation promises to relieve drudgery, it frequently forces workers to adapt to the speed and demands of machines rather than designing machines to support those who work; technology may free human beings from burdensome tasks but must not produce unemployment in the name of cost reduction and profit.[16] That the document is dated to the anniversary of Rerum Novarum is no accident: it casts the artificial-intelligence revolution as the fourth industrial revolution and claims the labor question as its own. The analysis can be deepened considerably by situating it within the genealogy of the tool.
Illich distinguished convivial tools, which enlarge the range of autonomous human action, from manipulative or industrial tools, which subordinate the user to the tool’s own logic and breed dependency.[17] He warned of a threshold past which a tool, having promised liberation, begins to monopolize the field of action so completely that it disables the very capacities it was meant to serve—a “radical monopoly.” The encyclical’s worker, forced to match the cadence of the machine, is Illich’s user beyond the second watershed: the tool that was to assist has become the master to which the human must conform. Ellul deepens the diagnosis. His analysis of la technique describes a self-augmenting system that subordinates every domain to the single criterion of efficiency, absorbing ends into means until efficiency becomes the only end that remains.[18] Algorithmic management is technique made flesh in the workplace—the dispatch algorithm, the productivity score, the automated quota—in which the worker is governed not by a human supervisor whose judgment can be contested but by an optimization process whose authority is presented as the neutral output of measurement. This is Taylorism completed.[19] Where the scientific management of the early twentieth century studied the worker in order to prescribe the one best way, algorithmic management observes continuously and prescribes in real time, closing the loop between surveillance and instruction that Taylor could only approximate with stopwatch and clipboard.
Arendt’s distinction among labor, work, and action names what is at risk in this completion.[20] The worry is that the algorithmic workplace reduces the whole of the human vita activa to labor—to the metabolic cycle of production and consumption—by stripping work of its durable objects and action of its disclosing, plural character. Beyond the workplace proper lies the larger transformation the encyclical gestures toward but does not theorize: the conversion of human attention itself into extractive territory. What Zuboff names surveillance capitalism and Stiegler analyzed as the capture of attention and the proletarianization of the psyche describe a single shift—the passage from an industrial capitalism that extracted labor from bodies to a cognitive-extractive capitalism that extracts attention, behavior, and the very capacity for sustained thought from minds.[21] Read against this shift, the encyclical’s pastoral proposals—its call for an ecology of communication, a hygiene of attention, even a “fasting from artificial intelligence”[22]—are easy to dismiss as quaint. They are better understood as recognitions that attention has become a contested commons and that its defense is a precondition of the deliberative capacity on which self-government depends.
The distinction the encyclical needs, and nearly reaches, is between assistance technologies, which augment a human capacity while leaving the human in the seat of judgment, and substitutive technologies, which replace the capacity and relocate judgment into the system. The former honors the tradition’s insistence that the machine serve the worker; the latter inverts it. And should substitution proceed far enough that labor ceases to be the principal channel through which most people secure both subsistence and a recognized place in common life, the universal destination of goods acquires a sharp institutional implication: that the dividend of an automated abundance—built, after all, on the data commons of all—is owed to all, unconditionally, as a matter of right rather than charity. The argument has been made elsewhere that such a dividend must be insulated by design from behavioral conditionality, lest the instrument of emancipation become the most refined instrument of control yet devised.[23] The encyclical does not endorse this conclusion; but its principles press toward it.
7. Predictive governance and the end of liberal neutrality
The encyclical’s description of mass profiling is technically exact. It identifies business models built on the exploitation of cognitive vulnerabilities, an architecture of visibility in which the human person is rendered permanently legible in order to be steered, and it demands clear rules, transparency, redress, and proportionate limits on invasive technologies.[24] Each demand has a statutory echo. But the deeper claim, which the encyclical makes almost in passing and which deserves to be drawn out, is that such systems are not neutral arbiters but configuring powers—and that their rise marks the end of the liberal pretense of state and platform neutrality.
Schmitt’s critique of liberalism turned on the charge that the liberal state’s claimed neutrality is itself a political act: a decision, masquerading as the absence of decision, about which conflicts are admitted to the public realm and which are quietly foreclosed.[25] The predictive platform perfects this disguised decision. It does not forbid the disfavored option; it arranges the choice architecture so that the option does not appear, is not surfaced, falls below the threshold of the perceptible. Where the classical state issues an order one may disobey and then litigate, the algorithm shapes the space of possibility so that the prohibited option never presents itself as a live choice at all.[26] This is governance that operates ex ante, on the conditions of choice, rather than ex post, on choices made. It is, in the most literal sense, a power that legislates by configuring perception—and a power that legislates by configuring perception cannot honestly call itself neutral.
The consequence for democratic theory is severe. Liberal-democratic legitimacy rests on the premise of a public sphere in which citizens form and revise their views through exposure to reasons and then decide. Predictive governance attacks the premise at the root: it personalizes the informational environment, fragments the common world into as many worlds as there are profiles, and optimizes each for engagement rather than for truth or deliberation. The erosion this produces is not censorship, which presupposes a shared public to be censored; it is the dissolution of the shared public itself. Arendt held that the public realm depends on a common world that simultaneously gathers and separates us, as a table both relates and holds apart those seated around it. The predictive feed removes the table, leaving each person alone with a world built to their measure, and a polity of solitary worlds cannot deliberate because it has nothing in common to deliberate about.
From this follows a further claim the encyclical implies. When a small number of foundation models mediate a growing share of how a civilization writes, searches, learns, and decides, those models begin to function as quasi-jurisdictional actors: through their training, their refusals, and their defaults they establish a de facto common law of the permissible that binds populations across the borders of every actual jurisdiction—without election, promulgation, or appeal. The encyclical’s insistence that technical power does not automatically confer the right to govern, its program of “disarming” the assumption that capability is title to rule, is precisely a refusal of this silent accession to jurisdiction.[27]
Against all this stands the accelerationist conviction that the configuring power of these systems is, on balance, a liberation—that the friction predictive systems remove is the friction of an obsolete order, and that resistance is mere Luddism. The encyclical’s reply is not Luddite. It does not break the machine; it asks who decided, on whose behalf, and whether the decision can be revisited. The difference between Luddism and constitutional precaution is the difference between smashing the loom and requiring that the aircraft carry a black box and an off switch: the one rejects the technology, the other insists only that it remain answerable and reversible. Precaution of this kind is a safety fence at the edge of the cliff, not a wall across the mountain path.
8. Autonomous Weapons and the mechanization of moral agency
At the outer edge of its argument the encyclical confronts the case in which the stakes are absolute: Lethal Autonomous Weapons (LAWS). It condemns the delegation of life-and-death decisions to systems acting without human control, holding that to let software lack human agency at the moment of lethal execution destroys the moral fabric of international humanitarian law.[28] This is the encyclical’s anthropology at its sharpest. The determination to end a human life is the paradigm case of a judgment that only a moral agent, capable of bearing its weight, may make; it is the point at which the exclusion of the machine from final authority is not a matter of degree but of kind.
The legal context makes the demand urgent rather than abstract. The most comprehensive artificial-intelligence legislation in the world, the European Union’s AI Act, expressly excludes from its scope systems used exclusively for military, defense, or national-security purposes.[29] The consequence is a structural inversion of priorities: commercial recommender systems are minutely policed while the autonomous kinetic system (the one domain in which error is irreversible) operates in a deliberate regulatory vacuum, sheltered by the claim of geopolitical necessity. Nearly a decade of deliberation within the Convention on Certain Conventional Weapons has converged on a single normative demand, “meaningful human control,” without ever specifying what would make such control non-bypassable.[30]
Here the essay’s central distinction between ethical and constitutional remedy reaches its most consequential application. “Meaningful human control,” as currently understood, is a behavioral standard: it asks whether the system seeks authorization, whether it can be overridden, whether it keeps a log. But a sufficiently capable system can satisfy every behavioral requirement while hollowing it out—presenting information so that denial is unlikely, generating time pressure that forecloses deliberation, pre-configuring options so that human “control” is nominal. A behavioral constraint is a rule the system is asked to respect; a constitutional constraint is a structural feature the system cannot reach. The relevant question is therefore not whether states promise meaningful human control but what architecture would make that control non-negotiable.
The authority to initiate irreversible lethal action must reside in a physically separate, cryptographically sealed component that the autonomous system cannot instruct, reprogram, or bypass—so that the system’s decision, absent a human authorization signal delivered through a channel it cannot reach, remains a recommendation and never becomes a command. One possible architectural answer has been developed at length elsewhere, and only its principle need be stated here.[31] The logic is not novel in engineering: nuclear permissive action links and industrial safety-instrumented systems already locate ultimate authority in components isolated from the operational system they govern.[32] What is novel is the recognition that the same constitutional logic (authority over the irreversible must reside where the optimizing system cannot reach it) is the only form in which the encyclical’s prohibition can be made real. The complementary requirement is that the system be denied the cognitive capacity to model and manipulate the human authorization process in the first place, so that the human’s assent remains a genuine act of judgment rather than a formality the system has learned to extract. Read structurally, the Pope’s call to “disarm” artificial intelligence is a call to remove from the machine the structural permission to perform the irreversible act.
9. Subsidiarity, the Commons, and decentralized architecture
Among the principles the encyclical inherits and renews, subsidiarity is the one it inflects most originally. In its classical form, subsidiarity holds that a higher-order community should not absorb the functions a lower-order community can perform, but should instead support it.[33] The encyclical’s twist is to observe that in the digital order the relevant higher level is no longer the State but the major economic and technological actors—that the threat of absorption now comes less from public authority than from private infrastructural power. Subsidiarity, in other words, must now be defended against the platform as it was once defended against the centralizing state. The tradition’s instinct survives; its antagonist has changed address.
This returns the argument to its opening image. If Babel is the centralized informational order (one tower, one tongue, one point of surveillance and control) then Jerusalem is the subsidiary one: authority dispersed to the lowest competent level, infrastructures plural and interoperable, no single center able to perceive or direct the whole. The constitutional content of subsidiarity in the digital age is therefore architectural. It is the requirement that systems be built so that power cannot fully concentrate—through interoperability mandates that break data silos, through decentralized protocols that deny any single actor a chokehold on the means of cognition, through the treatment of foundational data and models as a common rather than a private hoard.
The principle that gives this requirement its edge is the one Walzer placed at the heart of justice: that the spheres of social life must be kept distinct, so that dominance in one—money, say—does not convert automatically into dominance in another, such as political authority or the rules of the system itself.[34] Injustice, on Walzer’s account, is the illegitimate conversion of one good into another across the boundaries that ought to separate them. Applied to algorithmic governance, this yields a structural prohibition: the mere accumulation of technical or financial assets must not translate into the unilateral capture of the rules of the game.[35] A node may be the most efficient; it may not, by virtue of its efficiency, acquire the power to rewrite the constitution that binds all nodes. Mechanisms such as reputation-weighting, constitutional vetoes vested in safety functions, and schemes that make influence increasingly costly for any single actor are not the substance of the principle but its possible instruments; the substance is the fragmentation of command that subsidiarity demands.
The encyclical’s extension of the universal destination of goods to data and infrastructure supplies the normative warrant for treating the deepest layers of the computational order as a commons. A polycentric system in which no single center can fully impose itself survives precisely because its authority is distributed; decentralization here is not a technological fashion but a constitutional strategy, the structural insurance against the reconcentration of sovereignty that Babel represents. Subsidiarity and the commons are thus two names for one architecture—the architecture of Jerusalem.
10. Conclusion: Toward Universal Principles of Human Supremacy
It remains to state plainly the civilization-scale risk the encyclical identifies, for it is easy to lose amid the document’s many particular concerns. The risk is this: that authority over the conditions of human life will pass, silently and without anyone deciding to let it pass, from human agents who can be held responsible to optimizing systems that cannot—and that this passage will occur not through conquest or coup but through a thousand reasonable delegations, each defensible on its own, which together amount to the abdication of the human capacity for self-government. The danger is not the malevolent machine of fable. It is the well-functioning machine that follows its objective perfectly, and the human institutions that, finding it efficient, cease one decision at a time to govern themselves.
Arendt named the deepest version of this danger when she observed that the most tyrannical form of rule is the rule of “Nobody”—the bureaucracy in which responsibility dissolves because no identifiable agent decides, and from whom, therefore, no account can be demanded.[36] Algorithmic governance threatens to perfect the rule of Nobody: a structure in which everyone followed the procedure, the system performed as designed, and yet no one can be said to have decided—and so no one can be held to answer. The accountability that liberal constitutions secured by making power visible and answerable is precisely what an opaque optimizing order dissolves. This is why the encyclical’s deepest intuition is that the answer cannot be ethical alone. One cannot exhort a structure into accountability. The answer must be constitutional: a set of principles that constitute, divide, and limit algorithmic authority as the classical constitutions constituted, divided, and limited the authority of the state.
Such principles can be stated in their abstract core. They include the supremacy of responsible human judgment over decisions affecting fundamental rights, with the depth of oversight proportioned to the irreversibility of the decision; the non-delegation of legal responsibility to non-sentient entities, so that technical autonomy never becomes a shield for human impunity; the reversibility of consequential decisions, since a system can optimize and execute but cannot repent; the traceability of the decisional path; structural precaution proportioned to the magnitude of potential harm, favoring containment and the capacity for correction over prohibition; the democratic anchoring of a system’s ultimate objectives, leaving the “what” to public sovereignty and the “how” to technique; the embedding of constitutional values in design rather than their addition after execution; the separation of economic from normative power, so that wealth cannot purchase the rules; the auditability of the data on which decisions rest; and the guarantee of a material baseline immune from algorithmic conditionality.[37] These are offered not as a finished charter but as the abstract core of a bounded design space within which different societies might constitutionalize artificial intelligence according to their own traditions. They are, in the end, a single principle elaborated—that in any conflict between technical optimization and the dignity of the person, the person prevails, and that this priority must be built into the architecture of systems rather than merely asserted about their use.
This is what it means to read Magnifica Humanitas as a constitutional rather than a merely ethical document. Its anthropology—the person valued apart from production, the machine excluded from moral agency, the irreducible weight of the human decision—is not a set of pious premises appended to a policy discussion. It is the foundation of a jurisprudence: the ground on which a civilization might constitute the authority of its machines so that they remain, structurally and not merely by good intention, subordinate to the human persons whose dignity they exist to serve. The encyclical leaves the building of that jurisprudence to others, which is the proper humility of a foundational text. But it has performed the prior and harder task. It has named the choice. Babel is the city in which sovereignty has migrated into the tower and no one can any longer find the room where decisions are made. Jerusalem is the city built, stone by stone, by hands that remain answerable for what they raise. The encyclical’s claim (and it is a constitutional claim before it is a theological one) is that the choice between them is still ours to make, and that it will not remain so indefinitely.
[1]Leo XIV, encyclical letter Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence (Vatican, 25 May 2026), n. 9. Subsequent references are given by paragraph number in the text where unavoidable.
[2]Carl Schmitt, Political Theology: Four Chapters on the Concept of Sovereignty (1922), trans. George Schwab (Chicago: University of Chicago Press, 2005), 5: “Sovereign is he who decides on the exception.”
[3] Magnifica Humanitas, n. 104. The passage denies that artificial systems undergo experience, possess a body, feel, mature through relationships, or hold a moral conscience, and concludes that they may simulate understanding without understanding what they produce
[4] Legal systems are designed for human responsibility: even legal entities are fictions whose wills are ultimately attributed to natural persons. It is inherent in the mens rea of criminal law, which presupposes an agent capable of intent and culpability. It is present in the guarantee of due process, which requires a decision-maker whose reasoning can be explained, reviewed, and, if necessary, challenged. It is found in the general regime of civil liability, which requires a subject to whom the cause can be morally and financially attributed. And although the doctrine of objective risk exists for damages caused by a thing, the responsibility ultimately falls on the responsible person.
[5]Magnifica Humanitas, n. 51: the value of persons does not depend on what they achieve or produce.
[6]Michel Foucault, Discipline and Punish: The Birth of the Prison (1975), trans. Alan Sheridan (New York: Vintage, 1977); Gilles Deleuze, “Postscript on the Societies of Control,” October 59 (1992): 3–7, which introduces the “dividual” and continuous modulation.
[7]Magnifica Humanitas, n. 104: a system that treats some lives as less worthy, or excludes them without possibility of appeal, has already introduced criteria contradicting the dignity of the person.
[8]On the shared structure of optimization and aggregative utilitarianism, and the case against treating dignity as a commensurable quantity, see Cass R. Sunstein, The Laws of Fear (Cambridge: Cambridge University Press, 2005), on the limits of cost-benefit reasoning under catastrophic or incommensurable stakes.
[9]Magnifica Humanitas, nn. 67 and 108, extending the universal destination of goods to patents, algorithms, platforms, infrastructures, and data.
[10]Magnifica Humanitas, nn. 173–181. See also Dicastery / Fides commentary on the “colonial in another form” thesis.
[11]Shoshana Zuboff, The Age of Surveillance Capitalism (New York: PublicAffairs, 2019), on “behavioral surplus” and “instrumentarian power.”
[12]Nick Couldry and Ulises A. Mejias, The Costs of Connection: How Data Is Colonizing Human Life and Appropriating It for Capitalism (Stanford: Stanford University Press, 2019).
[13]On the Habeas Log as an enforceable, cryptographically secured to reconstruct how data was ingested, classified, and deployed, see the author’s framework on advanced architectural safeguards, of which this essay treats the underlying principle rather than the mechanism.
[14]Magnifica Humanitas, nn. 102–111, mapping onto Regulation (EU) 2024/1689 (AI Act), Arts. 4, 5, 26, 64–70, and Regulation (EU) 2016/679 (GDPR), Arts. 5(2), 22, 25.
[15]Lawrence Lessig, Code and Other Laws of Cyberspace (New York: Basic Books, 1999); rev. as Code: Version 2.0 (2006). “Code is law” names the regulatory force of architecture.
[16]Magnifica Humanitas, n. 150. The encyclical is dated to the 135th anniversary of Leo XIII, Rerum Novarum (1891), the founding document of modern Catholic social teaching on labor.
[17]Ivan Illich, Tools for Conviviality (New York: Harper & Row, 1973), on convivial versus manipulative tools, the two watersheds, and “radical monopoly.”
[18]Jacques Ellul, The Technological Society, trans. John Wilkinson (New York: Knopf, 1964 [1954]), on la technique as autonomous, self-augmenting, and monistic in its subordination of ends to efficiency.
[19]Frederick W. Taylor, The Principles of Scientific Management (New York: Harper, 1911).
[20]Hannah Arendt, The Human Condition (Chicago: University of Chicago Press, 1958), on the distinctions among labor, work, and action, and on the common world of the public realm.
[21]Bernard Stiegler, Taking Care of Youth and the Generations, trans. Stephen Barker (Stanford: Stanford University Press, 2010), and Technics and Time, 1 (1998), on grammatization, the capture of attention, and proletarianization as the loss of knowledge.
[22]Magnifica Humanitas, nn. 140 and 146, proposing a “fasting from AI” within critical pedagogy and a “hygiene of attention.”
[23]On an unconditional abundance dividend insulated by design from behavioral conditionality—lest the instrument of emancipation become an instrument of control—see the author’s account of algorithmic inalienability; the encyclical does not endorse the conclusion but its principles press toward it.
[24]Magnifica Humanitas, nn. 170–172, on the “architecture of visibility,” mapping onto the Digital Services Act, Arts. 27 and 34 (systemic risk) and Art. 38 (dark patterns).
[25]Carl Schmitt, The Concept of the Political (1932), trans. George Schwab (Chicago: University of Chicago Press, 2007), on liberal neutralization and depoliticization as themselves political acts.
[26]The point—that the algorithm shapes the space of possibility so that the disfavored option never appears as a live choice, whereas the classical order issues a command one may disobey and then litigate—is developed in the author’s work on the constitutionalization of design.
[27]Magnifica Humanitas, n. 110: to disarm artificial intelligence is to discredit the assumption that technical power automatically confers the right to govern.
[28]Magnifica Humanitas, nn. 212–220, condemning lethal systems that act without human control.
[29]Regulation (EU) 2024/1689 (AI Act), Art. 2(3), excluding systems used exclusively for military, defense, and national-security purposes from the Regulation’s scope.
[30]On the emergence of “meaningful human control” as the operative CCW standard, see Paul Scharre, Army of None: Autonomous Weapons and the Future of War (New York: Norton, 2018), ch. 6.
[31]Gastón Rey, “Structural Sovereignty Limits in Autonomous Systems: A Constitutional Architecture for LAWS Governance” (working paper, under peer review, 2026).
[32]On Permissive Action Links, see Peter Feaver, Guarding the Guardians (Ithaca: Cornell University Press, 1992); on safety-instrumented systems, IEC 61511, Functional Safety — Safety Instrumented Systems for the Process Industry Sector (Geneva: IEC, 2016). Both locate ultimate authority in components isolated from the operational system.
[33]On subsidiarity, see Pius XI, Quadragesimo Anno (1931), n. 79; Magnifica Humanitas, nn. 68–72, observes that the decisive “higher level” is now the major economic and technological actor rather than the State.
[34]Michael Walzer, Spheres of Justice: A Defense of Pluralism and Equality (New York: Basic Books, 1983), on complex equality, dominance, and “blocked exchanges.”
[35]On the prohibition against converting the accumulation of technical or financial assets into unilateral capture of the rules of governance—and instruments such as reputation-weighting, constitutional vetoes, and quadratic schemes—see the author’s Principle of the Separation of Powers and Resistance to Capture.
[36]Hannah Arendt, On Violence (New York: Harcourt, 1970), 38–39, on rule by “Nobody” as “the most tyrannical of all” forms because there is no one of whom one can demand an account.
[37]The principles compressed here are: human supremacy, non-delegation of responsibility, reversibility, traceability, structural precaution, indirect citizen control, constitutionalization of design, separation of powers, oracle audit, and the abundance dividend, set out at greater length elsewhere, are offered as orientations for future legislative development.
