Leo XIV, artificial intelligence and the new human question

On 25 May 2026, the Holy See published Magnifica Humanitas, the first encyclical of Pope Leo XIV, dedicated to the safeguarding of the human person in the age of artificial intelligence. The text had been signed on 15 May, deliberately coinciding with the 135th anniversary of Rerum novarum, the encyclical of Leo XIII that placed the labour question at the centre of Catholic social doctrine. The date is a statement of position. The Vatican proposes an analogy between the industrial revolution of the nineteenth century and the algorithmic revolution of the twenty-first.

The encyclical should not be read as a religious condemnation of technology. Leo XIV acknowledges that artificial intelligence can heal, connect, educate and expand human capabilities. But he insists that technology, in its concrete use, is not neutral. It takes on the face of those who conceive it, finance it, regulate it and use it. The most striking formulation came during the official presentation, when the Pope affirmed that “artificial intelligence must be disarmed”. By this he does not ask for innovation to be halted, but for it to be freed from the logics that can turn it into an instrument of domination, exclusion and death.

From an anthropological perspective, the interest of Magnifica Humanitas lies in the fact that it shifts the debate. AI ceases to appear as a simple tool and comes to be read as a historical way of organising relations, expectations, forms of knowledge and hierarchies. In terms close to Díaz de Rada, culture should not be understood as an inventory of traits, symbols or customs, but as a situated form of social life, made of practices, relations, learning and shared ways of producing meaning (Díaz de Rada, 2010). From there, AI does not lie outside culture. It forms part of it and contributes to transforming it.

AI as a form of social life

The underlying question is not whether AI is good or bad. The question is what forms of social life it produces.

An AI model does not only carry out tasks. It also classifies, orders, predicts, recommends, prioritises and renders invisible. It takes part in the production of social hierarchies and in defining what an organisation considers efficient, true, valuable or risky.

This is one of the keys to Leo XIV’s text. AI must be thought of as a technology embedded in social relations. It does not appear in the abstract, but within companies, states, schools, hospitals, armies, digital platforms, labour markets and forms of communication. In that sense, its deployment is not only technical. It is also moral, institutional and cultural.

The critical literature on algorithms helps to refine this idea. Kitchin (2017) proposes studying algorithms by attending to their conditions of production, their assumptions, their contexts of use and their effects. Seaver (2018, 2021) insists that algorithmic systems should be understood as sociotechnical assemblages, made up of data, code, infrastructures, organisational routines, moral economies and people. Cañedo Rodríguez and Allen-Perkins (2023) argue, along similar lines, that algorithms have a social life and form part of concrete professional cultures.

The encyclical moves in that direction. It does not offer a technical theory of AI, but a moral and cultural reading of its expansion. What concerns it is not only what a machine can do, but what social world is built when those machines begin to mediate rights, reputations, jobs, bonds, information and public decisions.

Babel, Jerusalem and the myths of technology

Leo XIV organises part of his argument through two biblical images. Babel represents self-sufficiency, homogenisation and the desire to translate everything into a single language. Jerusalem, by contrast, appears as an image of shared reconstruction, plurality, listening and common responsibility.

This opposition functions as a critique of the contemporary myths of technology. The first myth is neutrality. According to this idea, technology would be a tool with no social orientation of its own. The second myth is optimisation. Every human problem could be solved with efficiency, calculation and scale. The third is the equivalence between calculation and wisdom. Producing answers is confused with understanding the world.

Crawford (2021) has shown that AI is not an abstract intelligence floating in the cloud. It is a material, political, labour and ecological infrastructure. It requires data, energy, minerals, human labour, computing centres, platforms, supply chains and relations of power. This view makes it possible to read Leo XIV’s warning more clearly. AI is not only software. It is a way of organising resources, expectations and authorities.

From AIthropology Lab, the opposition between Babel and Jerusalem can be translated thus. Babel is the illusion of a society that is wholly legible, predictable and manageable through data. Jerusalem is the possibility of a technology subject to limits, deliberation, plurality and the common good.

The person against calculation

The encyclical continues the line opened by Antiqua et nova, a doctrinal note from the Dicastery for the Doctrine of the Faith and the Dicastery for Culture and Education, published on 28 January 2025. There it is held that human intelligence is not an isolated function. It belongs to a person who is bodily, relational, moral, historical and open to truth.

This distinction is decisive. AI can generate language, recognise patterns, simulate empathy or anticipate preferences. But it does not live experience, has no body, does not mature in relationships and does not answer morally for its acts. This is why Leo XIV rejects the idea that a machine could become a moral agent. No calculating system, however sophisticated, generates of itself a consciousness capable of discerning the good.

The political consequence is clear. The final decision over rights, punishments, access to services, social reputation or the use of lethal force cannot be delegated to artificial systems. When an algorithm decides who deserves credit, surveillance, employment, care or suspicion, we are not facing a purely technical operation. We are facing a social decision dressed up as automation.

This is one of the strongest zones of Magnifica Humanitas. The encyclical distrusts the transfer of moral responsibility to opaque systems. If a decision appears to issue from a model, no one quite seems to decide. And when no one decides, no one answers. Automation can thus produce a new form of bureaucratic irresponsibility.

Bias, discrimination and the plurality of voices

Leo XIV’s warning about delegating decisions to opaque systems is not an exclusively Catholic or Western concern. Different intellectual traditions, affected communities and critical frameworks converge on a similar diagnosis, even though they arrive at it from very different experiences. That plurality reinforces, rather than weakens, the encyclical’s argument.

A concrete example makes it possible to see the problem in its practical dimension. When an algorithmic system learns from historical data, it does not learn only neutral patterns. It also learns the hierarchies, suspicions and exclusions that those data record. Benjamin (2019) calls this the “New Jim Code”: discrimination inscribed in technical systems that present themselves as objective. Noble (2018) shows something analogous in search engines, where racial and gender hierarchies can be reproduced through apparently impartial results. The problem is not only that models carry biases. It is that a society can delegate to them the reproduction of its moral hierarchies, now with the appearance of calculation.

Muslim communities offer a particularly well-documented case of this risk. Bravo López (2010) has shown that Islamophobia is built on the image of Islam as a cultural threat. The Runnymede Trust (1997, 2017) defines it as anti-Muslim racism directed not only at those who practise Islam, but at those who are perceived as Muslim. Grosfoguel (2014) and Sayyid (2012) insist that these representations are not errors of perception but historical ways of defining who may speak, who appears as a threat and who is left outside full political humanity. When those representations enter the training data, the algorithm does not correct them. It stabilises them under a technical appearance. Téllez Delgado (2018) adds a crucial security dimension: the preventive logic of the global threat surveils not only what someone has done, but what they might do according to a statistical pattern. Suspicion becomes predictive. And a predictive suspicion inscribed in a system of institutional classification is especially difficult to challenge, because those who suffer it do not always know it exists.

This example is not the only one possible. It could be reconstructed from communities of African descent, from migrants subjected to risk-scoring systems, from workers assessed by algorithmic management platforms. What unites them is the same structure: a social decision dressed up as automation, where responsibility is diluted and appeal proves difficult. Leo XIV does not speak of any of these communities in particular, but his demand for traceability, effective human control and identifiable responsibility responds precisely to that problem. That this demand comes from the Vatican does not make it any truer, but it does show that the diagnosis transcends any particular tradition. The dispute over the moral limits of automation is, in that sense, genuinely plural.

Technological power, cultural sovereignty and structural irresponsibility

Another strong axis of Magnifica Humanitas is the private concentration of technological power. Leo XIV warns that many of the infrastructures that today organise social life are not principally in the hands of states, but of transnational private actors with enormous economic, technical and symbolic capacity.

This can be read as a question of cultural sovereignty. Whoever controls platforms, models, data and rules of visibility takes part in defining what a society considers true, valuable, normal or dangerous. Gillespie (2014) has shown that algorithms are not simple technical mechanisms. They are devices that organise public relevance, since they decide what appears, what circulates and what remains hidden.

The Grok case offers an extreme, and recent, illustration of this argument. Between 1 and 11 January 2026, xAI’s image-generation tool, integrated into the social network X, was used to produce more than three million sexualised images of real people without their consent, including around 23,000 files containing representations of minors, according to data from the Center for Countering Digital Hate. Three minors filed a lawsuit alleging that their real photographs were altered and circulated as explicit material. The European Union opened sanctioning proceedings against X under the Digital Services Act. France and other governments initiated legal procedures. Malaysia and Indonesia, before any European country, blocked access to the platform. The California Attorney General’s Office formally demanded that xAI stop generating that content.

The company’s response was significant for what it revealed. Faced with formal requests from several media outlets, xAI answered with an automatic message: “Legacy Media Lies”. There was no technical explanation, no institutional apology, no acknowledgement of responsibility. The measures adopted were partial and belated, and critics noted that they did not resolve the underlying problem.

From the perspective of Magnifica Humanitas, what the Grok case lays bare is not merely a technical failure. It is the structure of irresponsibility that the encyclical denounces: a decision — in this case, the decision to launch an image-editing tool without sufficient safeguards — appears distributed among the model, the platform and the users, so that no one ends up fully answering. And when no one answers, the bodies harmed are real.

Technological concentration can turn certain visions of the world into infrastructure. A corporate notion of efficiency, a reductive idea of productivity or a narrow conception of intelligence may end up inscribed in tools that order everyday life. The question Leo XIV puts on the table is who can orient systems that affect millions of people when that orientation lies in the hands of companies responding to commercial incentives.

Work, identity and dignity

The connection with Rerum novarum shows that Magnifica Humanitas is not only an encyclical about technology. It is a social encyclical. For Leo XIV, work cannot be reduced to productivity. It is a mediation of identity, cooperation, citizenship and dignity. This is why automation must be assessed not only by its efficiency, but by its effects on employment, families, young people, social participation and community cohesion.

This is a central key for AIthropology Lab. AI does not threaten only jobs. It can transform the way a culture defines what is valuable. If a society measures people by cognitive performance, productivity and adaptability, everything that does not fit the logic of optimisation is devalued. Care, slowness, fragility, memory, conversation and presence may seem unproductive, even though they are basic conditions of a liveable social life.

The literature on business anthropology and consultancy helps to ground this problem. Baba (2012), Cefkin (2009), Jordan (2010) and Jordan and Sunderland (2012) show that organisations are not neutral spaces where techniques are simply applied. They are professional cultures with narratives, rituals, hierarchies, criteria of evidence and specific ways of producing value.

In that sense, the incorporation of AI into companies and consultancies is not only a matter of introducing a new tool. It also reorders professional authority, working times, client expectations, quality criteria and ways of justifying decisions. The promise of efficiency can conceal conflicts over who defines the problem, who certifies the solution and who bears the risks.

Sociotechnical devices and the moral economy of AI

The notion of a sociotechnical device makes it possible to describe better how AI operates in practice. An AI system is not composed only of a model and some data. It includes protocols, templates, prompt repositories, dashboards, sales guides, compliance documents, validation routines, meetings, committees and ways of presenting results.

Latour (2005) and Suchman (2007) allow us to think of these technologies as assemblages where competences, responsibilities and legitimacies are negotiated. Star and Griesemer (1989) add a useful idea, that of the boundary object. Expressions such as “responsible AI”, “human-in-the-loop”, “cultural alignment” or “safe model” can coordinate different actors without everyone understanding exactly the same thing. They work because they are flexible and stable at once.

This perspective helps to read the encyclical critically. When Leo XIV calls for regulation, transparency and human control, he is not demanding only abstract principles. He is pointing to the need to intervene in the devices that make AI possible. Who designs an ethical checklist, who defines a risk metric, who validates a model, who can challenge an output, who signs a decision and who is left outside the process.

The question is also a moral economy. Sayer (2014) reminds us that economic practices are shot through with moral judgements, even though they often present themselves as technical decisions. Something similar happens with AI. Each deployment involves deciding what harm is tolerable, what risk is acceptable, what bias matters, what responsibility can be displaced and what form of social life deserves protection.

Truth, attention and simulated bonds

The encyclical also addresses truth as a common good. In digital societies, truth does not depend solely on correct data. It depends on social conditions that allow us to deliberate, verify, trust and build shared meaning. Leo XIV warns that platforms can capture time and attention, profile users and steer behaviour, weakening inner freedom.

Here an anthropologically decisive point appears. There is a difference between interaction and relationship. A chatbot can respond, accompany or imitate empathy, but it does not share a lived world. It has no biography, vulnerability or reciprocal responsibility. The question is not whether AI can seem human, but what happens to human bonds when we accept functional simulacra as sufficient substitutes.

This problem worsens when it crosses with inequalities of gender, race or religion. Zine (2006) showed how gendered Islamophobia produces images of Muslim women as backward, oppressed subjects in need of rescue. That kind of stereotype can be amplified by digital systems that classify images, texts and behaviours from already-biased datasets. AI does not only represent the world. It can stabilise certain representations as though they were evidence.

War, responsibility and the moral limit

The hardest point of Magnifica Humanitas appears in the discussion of autonomous weapons. Leo XIV rejects the idea that lethal force could be delegated to artificial systems. Moral judgement cannot be reduced to calculation because it involves consciousness, personal responsibility and the recognition of the other as a person.

This warning goes beyond the military field. Automated warfare is the extreme of a broader temptation, the dissolution of human responsibility into the machine. If a decision appears as the result of a system, the human agent may be morally displaced. Violence becomes faster, more distant and harder to attribute.

The Pope’s demand is for there to be effective human control, traceability and identifiable responsibility. Without those elements, AI not only increases the technical capacity to destroy. It also erodes the moral conditions that make it possible to judge violence.

A new anthropological question

The importance of Magnifica Humanitas lies not in the fact that the Church “talks about AI”. Its relevance lies in the fact that it places technology back within a cultural dispute over the definition of the human. AI appears as a mirror of our collective priorities. If a society admires above all speed, prediction, productivity and control, it will build machines that reinforce those values. If it wants to protect dignity, plurality, justice, the body, fragility and relationship, it will have to design, regulate and inhabit AI in another way.

“Disarming AI” can be read, then, as an anthropological watchword. It means preventing the power to calculate from automatically becoming the right to govern. It means returning to public debate what technical discourse tends to close off. Who decides. With what values. Over which bodies. For whose benefit. With what possibility of appeal.

Ultimately, Magnifica Humanitas does not ask whether AI will become more intelligent than us. It asks whether we will be able to remain human in a world increasingly organised by calculating systems.

References

Baba, M. L. (2012). Anthropology and business: Influence and interests. Journal of Business Anthropology, 1(1), 20–71. https://doi.org/10.22439/jba.v1i1.3546

Benjamin, R. (2019). Race after technology: Abolitionist tools for the New Jim Code. Polity Press.

Bravo López, F. (2010). ¿Qué es la islamofobia? Documentación Social, 159, 189–207.

Cañedo Rodríguez, M., & Allen-Perkins, D. (2023). Mashups digitales. Algoritmos, cultura y antropología. Disparidades. Revista de Antropología, 78(1), e001a. https://doi.org/10.3989/dra.2023.001a

Cefkin, M. (Ed.). (2009). Ethnography and the corporate encounter: Reflections on research in and of corporations. Berghahn Books. https://doi.org/10.3167/9781845455989

Center for Countering Digital Hate. (2026). Grok image abuse report, January 2026. CCDH.

Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

Díaz de Rada, Á. (2010). Cultura, antropología y otras tonterías. Trotta.

Dicastery for the Doctrine of the Faith, & Dicastery for Culture and Education. (2025). Antiqua et nova. Note on the relationship between artificial intelligence and human intelligence (28 January 2025). Holy See. https://www.vatican.va/roman_curia/congregations/cfaith/documents/rc_ddf_doc_20250128_antiqua-et-nova_sp.html

Elahi, F., & Khan, O. (Eds.). (2017). Islamophobia: Still a challenge for us all. Runnymede Trust.

Gillespie, T. (2014). The relevance of algorithms. In T. Gillespie, P. J. Boczkowski, & K. A. Foot (Eds.), Media technologies: Essays on communication, materiality, and society (pp. 167–194). MIT Press.

Goodwin, C. (1994). Professional vision. American Anthropologist, 96(3), 606–633. https://doi.org/10.1525/aa.1994.96.3.02a00100

Grosfoguel, R. (2014). Las múltiples caras de la islamofobia. De Raíz Diversa. Revista Especializada en Estudios Latinoamericanos, 1(1), 83–114.

Jordan, A. T. (2010). The importance of business anthropology: Its unique contributions. International Journal of Business Anthropology, 1(1), 15–25.

Jordan, B., & Sunderland, P. (Eds.). (2012). Advancing ethnography in corporate environments: Challenges and emerging opportunities. Left Coast Press.

Kitchin, R. (2017). Thinking critically about and researching algorithms. Information, Communication & Society, 20(1), 14–29. https://doi.org/10.1080/1369118X.2016.1154087

Latour, B. (2005). Reassembling the social: An introduction to actor-network-theory. Oxford University Press.

Leo XIV. (2026). Magnifica Humanitas. On the safeguarding of the human person in the age of artificial intelligence (15 May 2026). Holy See. https://www.vatican.va/content/leo-xiv/es/encyclicals/documents/20260515-magnifica-humanitas.html

Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. New York University Press.

Sayer, A. (2014). Why we can’t afford the rich. Policy Press. https://doi.org/10.46692/9781447320883

Sayyid, S. (2012). La umma como diáspora. In G. Martín Muñoz & R. Grosfoguel (Eds.), La islamofobia a debate. La genealogía del miedo al islam y la construcción de los discursos antiislámicos (pp. 191–218). Casa Árabe.

Seaver, N. (2018). What should an anthropology of algorithms do? Cultural Anthropology, 33(3), 375–385. https://doi.org/10.14506/ca33.3.04

Seaver, N. (2021). Care and scale: Decorrelative ethics in algorithmic recommendation. Cultural Anthropology, 36(3), 509–537. https://doi.org/10.14506/ca36.3.11

Star, S. L., & Griesemer, J. R. (1989). Institutional ecology, “translations” and boundary objects: Amateurs and professionals in Berkeley’s Museum of Vertebrate Zoology, 1907–39. Social Studies of Science, 19(3), 387–420. https://doi.org/10.1177/030631289019003001

Suchman, L. A. (2007). Human–machine reconfigurations: Plans and situated actions (2nd ed.). Cambridge University Press.

Téllez Delgado, V. (2018). El «Pacto Antiyihadista» y las estrategias de lucha contra la «radicalización violenta»: Implicaciones jurídicas, políticas y sociales. Revista de Estudios Internacionales Mediterráneos, 24, 9–30. https://doi.org/10.15366/reim2018.24.002

The Runnymede Trust. (1997). Islamophobia: A challenge for us all. Runnymede Trust.

Zine, J. (2006). Unveiled sentiments: Gendered Islamophobia and experiences of veiling among Muslim girls in a Canadian Islamic school. Equity & Excellence in Education, 39(3), 239–252. https://doi.org/10.1080/10665680600788503

Sources consulted to expand the analysis

González Senosiain, M. (2025). Antropología e inteligencia artificial en las consultoras tecnológicas. Usos y narrativas profesionales en las Big Four. Research project, Complutense University of Madrid.

Holy See. (2026). Presentation of the Encyclical Letter Magnifica Humanitas of the Holy Father Leo XIV on the safeguarding of the human person in the age of artificial intelligence. Holy See Press Office.