Magnifica Humanitas: AI that is “opaque and unilateral” or “oriented toward the common good”: who chooses its values?

Can a more “moral” AI become politically dangerous? This apparent paradox lies at the heart of Magnifica Humanitas.

Leo XIV’s encyclical on artificial intelligence does not simply call for models to be regulated, deepfakes to be limited or employment to be protected in the age of AI, as Rerum Novarum did 135 years ago, at the time of the second industrial revolution. It asks a deeper question: who decides the values that structure systems which, every day, increasingly sort our requests, rank our information, filter our content, assess our risks, guide our choices and shape our relationships?

“A more moral AI is useless if that morality is decided by a handful of people.” (§107), the Pope states. Indeed, according to him, “we cannot simply invoke the moralization of the machine, what is called the ‘alignment’ of AI with human values, without having the courage to add one further condition: the possibility of debating the ethical code to be used, by subjecting it to shared criteria of social justice.”

In the vocabulary of Silicon Valley, alignment essentially refers to a security operation: ensuring that a model does not help build a weapon, does not manipulate its user, does not produce dangerous advice, in short, does not step outside the framework set by its designers. For Leo XIV, the question is who sets that framework. This is the strength of the encyclical: it does not treat AI as a simple tool, but as a structuring social power that “is not neutral, because it takes on the face of those who design it, finance it, regulate it and use it” (§9).

A power that “tends to become opaque and escape public control”

This reality is all the more acute as these systems become central: far from being a conversational gadget, generative AI increasingly shapes education, work, law, care, customer relations, cybersecurity, the media, propaganda, war and administrative decision-making. It is becoming an interface between the individual and knowledge, between the citizen and the public sphere, between the company and its risks, between the State and its civil and military decisions. Yet “every technical device implies choices and priorities: what it measures, what it ignores, what it optimizes, and the way it classifies people and situations” (§104). Indeed, AI does not merely transport data or calculate; it organizes the world according to criteria, an implicit definition of what matters — in short, an ontology, to use Palantir’s vocabulary.

“With great power comes great responsibility,” to borrow the words of Marlène Schiappa… or Spider-Man. A power that Leo XIV refuses to see distributed so unequally. He observes that the engines of technological development are now “private actors, often transnational, endowed with resources and intervention capacities greater than those of many governments” (§5), and that “when a power of such magnitude is concentrated in a few hands, it tends to become opaque and escape public control” (§95), even to become a fiefdom in the hands of tech lords.

In practical terms, those who control “platforms, infrastructures, data and computing power” set “the conditions of access, the rules of visibility and the possibilities for participation” (§95), and therefore also an increasing share of the power to say what will be visible, acceptable, dangerous, true or false.

“To disarm does not mean renouncing technology, but preventing it from dominating the human being”

This is where Magnifica Humanitas goes beyond the usual debates on AI bias. Biases are real, but they are only part of the problem: correcting a racist, sexist, violent, etc., bias, or a hallucination, is not enough if the entire system remains governed by opaque moral criteria defined by a handful of oligopolistic actors.

However, the encyclical does not target only this invisible layer, but also the very real uses of AI. Thus, the most discussed word in the text is probably “disarm,” which should not be reduced to autonomous weapons: “To disarm AI is to remove it from the logic of armed competition, which today is no longer only military, but also economic and cognitive.” (§110). Leo XIV clarifies his thinking: “To disarm does not mean renouncing technology, but preventing it from dominating the human being,” while “AI is already an environment in which we are immersed and a power with which we must reckon.”

This is the heart of the problem: a tool can be unplugged, whereas an environment is something we live in; it shapes our gestures, reflexes, expectations and vulnerabilities. It is a reality our readers know well: in cybersecurity, we know that a system is never merely an isolated application; it is made up of dependencies, identities, permissions, suppliers, data…

AI adds a cognitive and almost sensitive layer to this attack surface: the encyclical notes that AI can imitate “words of advice, empathy, friendship, love” (§100). This imitation can be useful, but it can also create “the illusion of being in a relationship with an authentic personal subject.” Beyond the “misrepresentation of the product,” the risk is that humans “lose the very desire to truly seek out the other.”

“It is essential that responsibilities be clearly defined”

One can read this passage as a spiritual meditation on loneliness, but in the cyber world it takes on an operational dimension: an AI capable of simulating attention, kindness, closeness or urgency becomes a tool for personalized manipulation. It “accelerates cybercrime,” emphasized Michaël De Laet, team leader at Europol’s European Cybercrime Centre (EC3), during the INCYBER Forum 2026. During a panel discussion, he mentioned AI agents capable of identifying targets on LinkedIn, writing spear-phishing messages in two minutes, and even initiating approach conversations… This is no longer merely automated fraud; the human relationship itself is becoming industrializable. The fake is no longer only a piece of content; it becomes a credible relationship.

To counter this, “it is essential that responsibilities be clearly defined at every stage: from those who design and program systems, to those who use them, or those who decide to entrust them with concrete choices” (§105), Leo XIV believes. In the general context of opacity surrounding AI, “responsibility becomes decisive, namely the possibility of identifying who must ‘give an account’ of decisions, justify them, control them and, if necessary, challenge them and repair the damage that follows from them.”

AI ethics therefore cannot be a communication gimmick: “adequate legal frameworks, independent oversight, user education and a policy that does not abdicate its duty” are needed (§106).

Tech players that exercise “de facto power over the conditions of community life”

A governable AI must be explainable, auditable, correctable, challengeable and attributable. An automated decision that affects work, credit, services, reputation or security cannot remain locked inside a black box. This concern mirrors that of cybersecurity: responsibility — moral or operational — does not exist without mapping dependencies and risks. One cannot govern what one cannot see.

In Magnifica Humanitas, this approach is reflected concretely through the application of the principle of subsidiarity to AI. Traditionally, this principle serves to protect individuals, families, municipalities or intermediary bodies from absorption by the State. Here, “the higher level is not the State, but each major economic and technological actor exercising de facto power over the conditions of community life” (§71). Leo XIV recommends that “these processes should not be imposed from above in an opaque and unilateral manner, but should be oriented toward the common good through transparency, responsibility and real forms of participation — independent controls, transparency on algorithms, equitable access to data, appeal mechanisms.”

These are all concerns that overlap with those of professionals seeking a healthy and safe digital world. The parallel with the DSA is illuminating. This regulation states that it seeks to combat “systemic risks” linked to freedom of expression, media pluralism, public safety and electoral processes, but its critics fear that it will strengthen the censorship power of platforms and European bodies. This is the dilemma the encyclical helps formulate: protecting truth is necessary, but delegating to platforms the definition of what can be published, what is true and what is false risks worsening the very problem one claims to solve.

Toward an “ecology of communication”

This question of truth is naturally central for an institution that claims to be the custodian of its revelation in the spiritual realm. For the secular world, however, Magnifica Humanitas is more measured: “Truth is a common good, not the property of those who hold power or visibility” (§137). Yet “disinformation was not born with AI, but today it finds in it a powerful multiplier” (§132), Leo XIV recalls. This concern echoes those raised during a panel discussion at the INCYBER Forum 2026 by Jérôme Nevicato, a reserve lieutenant colonel assigned to COMCYBER-MI, the French Interior Ministry’s Cyber Command: “We have truly entered an era of industrialized deception.”

This is a major danger, as “the search for truth is an essential element of democracy” (§132), a warning that might seem conventional. Yet this is not the case, because the encyclical does not propose an automated Ministry of Truth, but rather an “ecology of communication” (§137): transparency of amplification mechanisms, data protection, serious journalism, places for calm debate, critical education, fact-checking, “training in the correct and critical use of digital tools, AI…” In other words, truth depends on institutions, practices, trust and counter-powers.

In this respect, Magnifica Humanitas is more interesting than a simple pontifical text against the excesses of AI. It does not merely call for machines to be regulated; it asks who will have the right to define what they consider safe, true, just, human, dangerous, priority or acceptable. While it recognizes the contributions of AI, the encyclical refuses to entrust its morality to those who already own the models, the data, the platforms and the computing power.

The issue, then, is not whether AI will be moral or not, but whether its morality will be democratically debatable or privatized within the infrastructure. An AI aligned by a few can become a form of soft domination: polite, useful, efficient, reassuring, but structuring. A truly human AI should be something else: a habitable environment, that is to say, one that is understandable, contestable, governable and shareable. In pontifical language, “the MAGNIFICENT HUMANITY created by God now faces a decisive choice: to erect a new Tower of Babel or to build the city where God and humanity dwell together.” (§1)