I Come From Philosophy, and I Stayed an Atheist
I come from philosophy. Years when I read Garin and Cassirer in the morning and Foucault and Bobbio in the afternoon, where anti-clericalism and humanism coincided so naturally that you didn’t even need to argue it. From then on I have kept two convictions that I still consider the precondition of any serious public discourse. The first is that the historical religions are, in the vast majority of cases, anti-humanist devices, because the defence of the human requires a freedom that clerical structures have never been able to grant except through gritted teeth, under pressure, and almost always late. The second is that humanism, the real kind, the kind that starts with Garin and reaches the continental law of the twentieth century by way of the Italian Constitution and the 1948 Declaration, is a structurally secular ground, where religious traditions enter as guests, and when they try to take over the whole space they either get confused or betray themselves. These are judgments I have kept practising without much disturbance while reading the technical annexes of the AI Act, the latest implementing drafts of the Cyber Resilience Act, the EDPB guidance on DPIA templates for high-risk systems. I work in European compliance, I talk every day with public-administration and healthcare clients, and the only metaphysics I usually need is the already complicated one of Recital 71 of the GDPR. Then Magnifica Humanitas came out, and something moved.
Not an Exhortation, a Treatise on the Human
It took me two evenings to read it. It is not short, and it does not pretend to be. It is a long text, two hundred and forty-five paragraphs across five chapters, with an apparatus of two hundred and twenty-four footnotes, written in that curial prose that at times slows down and at times accelerates in an almost surprising way. I expected yet another moralising exhortation about artificial intelligence, the kind of document that tries to keep pace with an industry conference and ends up arriving two years late and half a thesis short. I found myself in front of something different. Not an exhortation, but a treatise. Not about AI, but about the human, with AI used as a touchstone to restate who we are. It is exactly the thing that secular technical debate struggles to produce, because it presupposes a level of conceptual foundation that we are used to delegating to the laws, hoping the laws will manage on their own.
Subsidiarity Was Not Born in Brussels
The problem is that the laws do not manage on their own. I see it clearly when I try to explain to a client why digital subsidiarity, the kind that inspires much of the architecture of the Data Act and the operational chapters of the AI Act, is not a Brussels invention but a principle with almost a century of elaboration behind it. Pius XI formulated it in 1931 in Quadragesimo Anno, and since then it has passed through case law, German ordoliberalism, Adenauer’s work, all the way to Article 5 of the Treaty on European Union. When a European regulator writes that certain decisions should be taken at the level closest to the people affected, it is using a grammar that was born elsewhere, not in Brussels. And when I try to defend that choice in front of a client, I find I lack the right words to explain it. Magnifica Humanitas, on this point, gave me back a vocabulary.
The Universal Destination of Goods, Applied to Algorithms
What struck me is not the novelty of the ideas. The encyclical’s ideas are largely known. What struck me is the way it holds them together, because they are exactly the ideas I need every time I have to discuss compliance without reducing it to a bureaucratic exercise. The universal destination of goods, for example, applied in paragraph 67 to patents, algorithms, platforms, infrastructure and data. It is not a position you can file away as pre-political or utopian. It is exactly the same problem the Data Governance Act tries to frame with the concept of data altruism, that the DMA chases with its gatekeeper rules, that the future interpretation of European health databases will have to face when the European Health Data Space comes into force. The fact that data is technically collectable and proprietarily assignable does not mean that its final destination is legitimately private. It is an operational statement, not a sermon.
Above the Citizen There Is No Longer Only the State
The same goes for subsidiarity in paragraph 71. The encyclical does something few technical documents have the courage to do explicitly: it recalls that, in the digital context, the level above the citizen is no longer the State, but the large economic actor that exercises de facto power over the conditions of common life. It is not a metaphor. When a platform decides what is visible and what is hidden, when an algorithmic scoring system establishes who gets credit and who does not, when a language model becomes the dominant interface for certain kinds of service, a power is exercised that the classical theory of subsidiarity had not foreseen in those terms. Recognising it formally, inside an official document of this authority, shifts the level of the discussion. It stops being a niche opinion. It becomes a grounded position that can be cited in a legal brief, in an impact assessment, in an opinion to a client who asks why all this effort is going into implementing a transparent logging system.
From Judgment to Calculation, Half a Century After Weizenbaum
Then there is a passage I did not expect, and it is probably the strongest piece of the whole document. It is in paragraph 198, in the chapter on the culture of power, where the encyclical addresses war, but it holds for everything. Moral judgment cannot be reduced to a calculation, because it involves conscience, personal responsibility and recognition of the other as a person. It is therefore not permissible to delegate lethal or otherwise irreversible decisions to artificial systems. Two sentences that contain at least half a century of critical thought on automatic calculation. Joseph Weizenbaum, the MIT engineer who in 1966 had built ELIZA and who a few years later was frightened to see how his secretary confided in it, wrote in 1976 a book whose subtitle was From Judgment to Calculation. The whole book was a defence of that preposition, to, which marks a loss. Weizenbaum argued that there are tasks computers cannot and must not perform, even if technically they could, because they involve the recognition of the other as a person, and that recognition is not a problem of internal representation. It is an act. Magnifica Humanitas, half a century later, says exactly the same thing with a different vocabulary. It is not a coincidence. It is that certain anthropological truths come back out when material conditions force them to come back out, and today the material conditions are those of automated decision-making applied to real human lives.
The New Slaveries and Data Colonialism
The encyclical goes further. In paragraph 173 it names things the mainstream debate on AI prefers to leave in the margins of vendor conferences. It explicitly cites data labeling, content moderation, rare-earth extraction, child labor in mines. It calls them new slaveries, and not as rhetoric. It calls them that because they are actually feeding, invisibly to anyone using an API at sixty cents per million tokens, the entire economy of transformer architectures. In paragraph 178 it introduces a concept that technopolitics scholars have been using for at least five years but that makes a certain impression when you see it used in a pontifical document: data colonialism. The key sentence is that whoever owns the health data of entire populations, today often collected under the sign of aid, research or innovation, in fact owns structural leverage on the future. It is an exact description of the strategic risk that weighs on many African, Asian and Latin American contexts, and that should weigh on our procurement choices when we decide where to host data, which vendor to buy a model’s training from, which ecosystem to delegate the epidemiological analysis of a region to. It is not a new-age sensibility. It is an impact assessment.
A Treatise That Holds as a Source for a DPIA
At this point in the reading I stopped. I was annotating passages in the margin as if I were preparing a technical brief, and I realised that was exactly what I was doing. The encyclical works perfectly well as an anthropological treatise, but it also works as a secondary source for a compliance argument. It is a text that can be cited in a DPIA to justify a restrictive assessment. It is a text that can be attached to an opinion on an AI integration in healthcare. Not because it has legal value, obviously, but because it provides the frame of principle that is often missing when technical choices are discussed, and that when missing turns into decisions made on the sole basis of opportunity cost. Anyone working in European compliance knows what I mean. There is a grey zone, between the recital and the article, between the legislator’s intent and the letter of the rule, where the practitioner needs to lean on something solid to explain why one interpretation is preferable to another. Having a text of this weight, written from a position outside the technical system but able to speak about it from the inside, is an asset it would be foolish not to use out of prejudice.
It Isn’t a Conversion, and It Isn’t a Perfect Text
I realise that anyone who knows me might find it strange to read these lines. I am not announcing a conversion, and I am not even saying the document is perfect. There are points where the encyclical slips, especially when it enters the ground of bioethics and the family, where my personal distance stays intact. There are passages where the pastoral tone takes over from conceptual precision, and where I would have wanted less rhetorical charity and more tools of analysis. It is not a text you sign, it is a text you argue with. But it is the kind of argument that I, as an atheist who works on the European regulation of technology, need today more than I needed yesterday. And I think it is needed, in the same way, by anyone trying to build a European technology industry different from the one that reaches us from San Francisco or Hangzhou, because different means founded on something, and something, at this historical stage, cannot be improvised.
Disarm
In paragraph 110 there is a word that made me smile. Disarm. The encyclical says it wants to disarm AI, and immediately specifies that to disarm does not mean to renounce. It means taking it out of the hands of monopolies and the logic of competition, making it debatable and contestable, returning it to the plurality of human cultures. It is a political word, not a religious one. It is exactly the word Brussels still cannot bring itself to say, because Brussels is still hostage to the realism that assumes the arms race as a natural given. A race that has stopped being military and become cognitive, as the encyclical notes with remarkable lucidity, but that keeps the same underlying logic. The fact that the first to say it, in an official document of this reach, was the Vatican and not a European commissioner, says something about the state of our public debate. It is not good news, but it is news to start from.
Three Things I Borrow
I close with a note that has nothing to do with theology. When you read a text produced inside a tradition that is not your own, there are two roads. One is preventive suspicion, which consists in asking the text to betray its own origin before you accept even to hear it. The other is the loan, which consists in taking what you need, leaving what you don’t, and going back to your work with one more tool. As an atheist, I feel I can borrow at least three things from Magnifica Humanitas. The explicit recognition that moral judgment cannot be automated, and that whoever tries to automate it is performing a political operation, not a technical one. The naming of the invisible supply chains that feed the digital economy, which is the precondition for any serious request for due diligence. And that word, disarm, which I would like to see applied, in the coming years, to the regulatory texts I will happen to read for work.
The rest, whoever cares, can argue in the proper places. For me, today, it is enough to have one more treatise on the nightstand, and a few well-written pages to reread when a client asks me why, after all, deep down, I am taking the trouble to do things properly.
Key takeaways
Magnifica Humanitas, Leo XIV’s first encyclical (signed 15 May 2026, the 135th anniversary of Rerum Novarum), is not an exhortation about AI but a treatise on the human, with AI used as a touchstone to restate who we are. Two hundred and forty-five paragraphs in five chapters, two hundred and twenty-four footnotes. It works as an anthropological treatise and as a secondary source for a compliance argument.
The universal destination of goods, applied in paragraph 67 to patents, algorithms, platforms, infrastructure and data, is the same question the Data Governance Act chases with data altruism, the DMA with its gatekeeper rules, the European Health Data Space with health databases. The fact that data is technically collectable and proprietarily assignable does not mean its final destination is legitimately private.
In paragraph 71 the encyclical recognises that, in the digital sphere, the level above the citizen is no longer the State but the large economic actor that exercises de facto power over the conditions of common life. It is the theory of subsidiarity updated to the power of platforms, a grounded position you can cite in a legal brief or an impact assessment.
Moral judgment cannot be reduced to a calculation (paragraph 198). It is the same thesis Joseph Weizenbaum defended in 1976 with the subtitle From Judgment to Calculation, half a century earlier. Certain anthropological truths come back when material conditions force them to, and today the material conditions are those of automated decision-making applied to real human lives.
In paragraphs 173 and 178 the encyclical names the invisible supply chains that feed the digital economy (data labeling, content moderation, rare-earth extraction, child labor in mines) and introduces data colonialism. Whoever owns the health data of entire populations owns structural leverage on the future. This is not new-age sensibility, it is an impact assessment.
As an atheist, I borrow three things. The recognition that moral judgment cannot be automated, and that whoever tries to automate it is performing a political operation, not a technical one. The naming of the invisible supply chains, the precondition for any serious due diligence. And the word disarm (paragraph 110), which does not mean to renounce, and which Brussels still cannot bring itself to say.
Questions & answers
What is Magnifica Humanitas?
It is Pope Leo XIV’s first encyclical, on the protection of the human person in the time of artificial intelligence. It was signed on 15 May 2026, the 135th anniversary of the promulgation of Leo XIII’s Rerum Novarum, and presented publicly on 25 May. It is a long text, two hundred and forty-five paragraphs across five chapters, with two hundred and twenty-four footnotes. For the first time, a Pope took part in person in the presentation of his own encyclical.
As an atheist, why would an encyclical interest you professionally?
Because anyone working in European compliance knows the grey zone between the recital and the article, between the legislator’s intent and the letter of the rule, where the practitioner needs something solid to lean on to explain why one interpretation is preferable to another. The encyclical has no legal value, of course, but it provides the frame of principle that is often missing when technical choices are discussed, and that when missing turns into decisions made on the sole basis of opportunity cost. It is a text that can be cited in a DPIA to justify a restrictive assessment. It is an asset it would be foolish to skip out of prejudice.
What does Catholic social doctrine have to do with European digital law?
More than you would think. Subsidiarity, which inspires much of the architecture of the Data Act and the operational chapters of the AI Act, is not a Brussels invention. Pius XI formulated it in 1931 in Quadragesimo Anno, and since then it has passed through case law, German ordoliberalism, Adenauer’s work, all the way to Article 5 of the Treaty on European Union. When a European regulator writes that certain decisions should be taken at the level closest to the people affected, it is using a grammar born elsewhere.
What does the encyclical say about automated judgment?
In paragraph 198 it argues that moral judgment cannot be reduced to a calculation, because it involves conscience, personal responsibility and recognition of the other as a person, and that it is therefore not permissible to delegate lethal or otherwise irreversible decisions to artificial systems. It is the same thesis Joseph Weizenbaum, the MIT engineer who built ELIZA in 1966, defended in 1976 in the book whose subtitle was From Judgment to Calculation. From judgment to calculation, a preposition that marks a loss.
What does it mean to disarm AI?
In paragraph 110 the encyclical says it wants to disarm AI, and immediately specifies that to disarm does not mean to renounce. It means taking AI out of the hands of monopolies and the logic of competition, making it debatable and contestable, returning it to the plurality of human cultures. It is a political word, not a religious one. It is exactly the word Brussels still cannot bring itself to say, because Brussels is still hostage to the realism that assumes the arms race, now cognitive more than military, as a natural given.
Is this a conversion?
No. I am not announcing a conversion and I am not saying the document is perfect. There are points where the encyclical slips, especially on bioethics and the family, where my personal distance stays intact, and passages where the pastoral tone takes over from conceptual precision. It is not a text you sign, it is a text you argue with. When you read a text produced inside a tradition that is not your own, you can choose preventive suspicion or you can choose the loan. I choose the loan.
The author
