The Encyclical and the Lab
How "Magnifica Humanitas" built the most serious AI framework to date — and what Olah's speech in the Synod Hall exposed
Magnifica Humanitas is the most substantive institutional engagement with artificial intelligence yet produced in the encyclical form — not because it says things no one has said before about AI ethics, but because it grounds the entire question in a philosophical and theological architecture of sufficient depth to resist the instrumental logic it diagnoses. Most coverage of Pope Leo XIV’s first encyclical has focused on the headlines: the Pope warns about AI, calls for regulation, urges caution. That is not wrong, but it misses what makes this document different from a white paper or a policy brief. The encyclical does not treat AI as a technical problem requiring governance solutions. It treats AI as a question about what kind of beings we are — and it answers that question with a rigour that the tech industry has not matched and most commentary has not noticed.
First, a confession is in order: I am not a theologian, and certainly not a specialist in contemporary Catholic doctrine. What follows is a close reading, not an authoritative theological analysis — an attempt to identify what the encyclical is actually arguing, where its arguments are unespectedly strong, and where they encounter their limits. Sure enough, any interpretations I make carry no confessional authority, and theologians may read the relevant passages differently. No definitive claims here, merely an invitation to further discussion
The document’s scope is wider than the AI question alone. This is a comprehensive restatement of Catholic Social Teaching for the digital age, and some of its most remarkable moves have little to do with technology as such. It declares the traditional “just war” theory — the framework through which the Church has evaluated military action since Augustine — outdated, at least as applied to the new forms of warfare the encyclical identifies: hybrid conflict, cyberattacks, and autonomous weapons systems. Whether this constitutes a development of just war theory or a repudiation of it is a question theologians will probably contest. It issues a formal apology for the Church’s institutional complicity in slavery, pointing out the Apostolic See’s own role in legitimating enslavement for eighteen centuries. Moreover, demands that the same standards of accountability it applies to tech companies and governments apply within the Church’s own structures. These are not tangential observations; they establish the kind of document this is — one that is willing to subject its own institution to the moral framework it proposes for others. The AI analysis draws its credibility from this willingness.
What gives the AI engagement its depth is that it finds itself within a tradition that has been developing, revising itself, and admitting errors for over a century now. The encyclical’s first two chapters make this point explicitly: Catholic Social Teaching is a living framework rather than a fixed set of rules, — and that is how the document authorizes itself to speak on a question that did not exist a decade ago. Development is the tradition here.
On the day of the encyclical’s release, Christopher Olah, co-founder of Anthropic, spoke in the Synod Hall at the Vatican. Anthropic is the company that has gone further than others in trying to do what the encyclical demands. It has published a Constitution for its AI system, Claude — an explicit set of values loosely governing Claude’s behaviour. It funds an advanced research programme in mechanistic interpretability — the effort to understand what is actually happening inside Large Language Models. And it introduced a Responsible Scaling Policy that committed the company to pause development if safety measures proved inadequate — a commitment that was, however, restructured in February 2026 in order to remove its binding character. For now, Olah’s speech is the most revealing response the encyclical has received — not because it is comparable in scope, but because of what it confirms, what it destabilizes, and what it conspicuously avoids. I will return to it after giving the encyclical the attention it deserves.
What Kind of Thing Is AI?
Somewhat surprisingly, the encyclical’s analysis of artificial intelligence begins not with ethics but with epistemology — with the question of what we actually know about the systems we are building.
§98 makes a technically precise observation that most commentary has underplayed: AI systems are more “cultivated” than “built.” Developers do not design every detail; they create a framework within which intelligence “grows.” The encyclical is careful here. It is not making a mystical claim about machine consciousness. It is making a practical point: the people who train these systems do not fully understand how they work. This is not a controversial statement — it is close to a consensus view among AI researchers — but it is unusual to see it stated this plainly in a document of this kind, and the encyclical draws a specific conclusion from it. If we do not fully understand what we have made, and we are deploying it at scale in consequential decision-making processes — employment, credit, public services, criminal justice — then the gap between our understanding and our deployment demands an unusual degree of institutional humility. Slower pace, more oversight, more transparency.
This is not a Luddite argument. The encyclical explicitly says technology has improved human life across centuries and is “a profoundly human reality, linked to the autonomy and freedom of man.” The argument is about proportion: the speed and scale of deployment relative to the depth of understanding. Humans routinely cultivate things they do not fully control — children, ecosystems, institutions. What makes AI distinctive in the encyclical’s analysis is that we are embedding something we do not yet adequately comprehend into the structures that govern people’s lives.
From this epistemological foundation, the encyclical moves to a claim that should make the AI industry uncomfortable.
§104 rejects the idea that AI can be considered morally neutral. The argument is not merely that tools can be used for good or bad purposes — that much is obvious. It is that every technical tool “embodies choices and priorities through what it measures, ignores and optimizes, and how it classifies people and situations.” The design process itself already contains a vision of what a human being is and what matters about human life. If a system is designed in a way that treats some lives as less worthy, or excludes people without the possibility of appeal, then the problem is not in how the tool is being used — the problem is already in how it was made. Ethical discernment, the encyclical argues, must examine not just purposes but design: “what vision of the human person and society is embedded in the data and models.”
This sounds abstract until you think about a credit-scoring algorithm that treats a postal code as a proxy for reliability, or a hiring tool that filters out résumés with employment gaps. These systems are not “neutral” instruments being applied badly. They have already made moral decisions — about what counts as creditworthiness, about what a good employee looks like — and those decisions are embedded in the architecture before any user touches them. The encyclical’s §103 puts this sharply: when automated systems make consequential decisions, “the exclusion of the vulnerable becomes cloaked in a veneer of neutrality and objectivity, against which it becomes difficult to raise objections.”
§107 pushes this further, and here the encyclical says something that should unsettle the AI safety community specifically. It is not enough, Leo XIV argues, to call for “the moralization of machines — the so-called ‘alignment’ of AI with human values” — without insisting on a further condition: the possibility of openly discussing the ethical frameworks involved and subjecting them to shared standards of social justice. “Otherwise, those who control AI will impose their own moral vision, which will become the invisible infrastructure of these systems.”
Read this carefully. The encyclical is not dismissing alignment work. It is naming a question the alignment community has been aware of but has not resolved: aligned to whose values? Decided how? With what democratic accountability? When a company like Anthropic or OpenAI decides what its model will refuse, what it will recommend, what tone it will take on contested moral questions — it is making moral law for potentially billions of users. The encyclical’s point is that these decisions are currently private, made by small teams within commercial organizations, with no democratic mandate and limited public scrutiny. It calls this “invisible infrastructure.” A more moral AI, the Pope writes, “is not enough if that morality is determined by a few.”
Here the encyclical most directly addresses the AI industry’s self-understanding — and sets up a tension that will return when we turn to Olah’s speech: the company he co-founded has gone further than any other in making its alignment decisions explicit and public — Claude’s Constitution is a published document. Whether that transparency is sufficient to answer the encyclical’s challenge, or whether it merely illustrates how far even the best current practice falls short, is one of the questions this document leaves open.
Personhood, Limitation, Incarnation
Secular readers will be tempted to skim what comes next. That would be a mistake. Beneath the governance recommendations and the calls for transparency lies an argument about what a human being is — and why that question, more than any regulatory scheme, determines whether we get AI right. The anthropological argument is doing the heaviest lifting in the entire document, and it is more rigorous than it might initially appear.
The argument starts with the Trinity. §48 locates the foundations of Catholic Social Teaching in “the mystery of the living God, revealed in Jesus Christ, who, as a communion of Persons — Father, Son and Holy Spirit — is love itself in relationship, expressed in the mutual gift of self.” §50 draws the consequence: every human person is “created for relationship,” planned and willed by God to enter into communion with others and with creation. Human dignity “does not depend on a person’s abilities, wealth or position in life” — it is a gift that precedes everything else.
For a non-theological reader, the key move here is what gets defined as the ground of personhood. It is not intelligence. It is not consciousness. It is not capability, output, or performance. It is relationship — specifically, the capacity for mutual self-gift, for vulnerability before another, for what the tradition calls communion. This is a philosophical position about the nature of being, not merely an ethical preference, and it has a direct consequence for AI: no matter how sophisticated a system’s outputs become, if it cannot enter into genuine mutual relationship — giving of itself, receiving from another, being changed by the encounter — then it is a fundamentally different kind of entity from a person. The question of whether AI “thinks” or “feels” becomes, in this framework, secondary. What matters is whether it can love and be loved, in the full theological sense.
The encyclical reinforces this with what I’d call its counter-programme to the logic of Silicon Valley. The Conclusion returns to the Incarnation — God becoming flesh, entering a body that suffers, ages, and dies. Christianity’s central claim is that the divine chose embodiment. The entire transhumanist project moves in the opposite direction: toward the overcoming of biological limits, toward uploading, optimization, the transcendence of the body. The encyclical is saying: from where we stand, that trajectory is not progress. It is a move away from the paradigm through which God chose to act in the world.
§§118–122 develop this into what is probably the most countercultural argument in the entire document. “Humanity flourishes not despite limitations, but often through them.” Everything that appears as a limit — illness, old age, suffering, vulnerability — “tends to be seen primarily as a defect to be corrected, rather than as a reality through which our humanity matures and opens itself to relationship.” In those moments of limitation, the encyclical says, people “discover a new wisdom, tangibly experience the closeness of others and encounter the presence of the Lord.”
This is not a celebration of suffering. The encyclical does not condemn medicine, prosthetics, or any technology that alleviates genuine human pain — these are works of mercy. The moral line it draws is between alleviating suffering, which is good, and the programmatic elimination of finitude as such — treating limitation itself as an error to be corrected — which is what the encyclical identifies as spiritually dangerous. The concern is that if you engineer away all vulnerability, you also engineer away the conditions under which compassion, dependence, generosity, and worship characteristically arise. The tradition has a deeper version of this argument: that genuine self-transcendence — becoming more than merely human — happens through grace and self-gift, not through computation. But even in its accessible form, the claim is striking: the drive to optimize away every human limitation is, for this document, not an expansion of human freedom but a contraction of human depth.
All of this builds toward §99, where the encyclical makes its most confident ontological claim about AI. It states that AI systems “do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean.” They “may imitate language, behavior and analytical skills, or even simulate empathy and understanding, but they do not understand what they produce.”
This is said with full conviction. The encyclical does not hedge or qualify. AI systems do not feel. They do not understand. They simulate. This is the foundation on which everything else — the moral framework, the governance recommendations, the insistence on human dignity — rests. If this claim is secure, then the encyclical’s entire architecture holds. If it is not, then questions open up that the document has not prepared itself to answer.
Power, Labour, Property
The encyclical is also a document about power — about who holds it in the digital age, how it operates, and who pays.
§71 makes a conceptual move that reorients Catholic political theology. Subsidiarity — the principle that decisions should be made at the lowest competent level, with higher authorities intervening only when necessary — has been a cornerstone of the Church’s social thought since Pius XI’s Quadragesimo Anno in 1931. But it was designed to protect individuals and communities from the overreach of states. Leo XIV argues that in the digital context, the “highest level” is no longer the state but “major economic and technological actors.” The effect is to reposition the entire tradition: the quasi-sovereign power against which subsidiarity must now be enforced is not a government but a handful of companies whose reach exceeds most governments’ and whose accountability falls far short.
The encyclical reads the culture that produced these companies through a specific theological lens. The Babel narrative — §7 maps the builders’ desire to “make a name for themselves” onto the founder-mythologies of the tech industry — is not deployed as a metaphor. In the encyclical’s framework, the telos of current AI development is not merely misguided but structurally idolatrous: it substitutes human self-assertion for the proper orientation of human activity, which is toward God and neighbour. This is a harder claim than “AI needs better ethics.” It is a claim about the spiritual character of an entire industrial culture.
The structural argument extends to competition. §110 introduces the verb disarmare applied to AI — framing the AI race as structurally analogous to arms proliferation, with the same dynamics of concentration, escalation, and mutual distrust that characterised nuclear competition. The argument connects to §192, where the encyclical declares the traditional “just war” theory — the framework the Church has used to evaluate military conflict since Augustine — “outdated.” The logic of the arms race, whether in weapons or in algorithms, is one this document refuses to sanctify.
On data and intellectual property, the encyclical makes a doctrinal move whose consequences have barely been discussed. §67 extends the principle of the “universal destination of goods” — the Catholic teaching that all goods are ultimately meant for the common benefit — to “patents, algorithms, digital platforms, technological infrastructure and data.” This principle has historically applied to land, resources, and the necessities of life. Extending it to algorithms and data means that, if the principle is taken at full strength, proprietary AI models and data hoarding carry the same moral weight as hoarding food while others starve. This is my reading of the doctrinal logic, not the encyclical’s explicit formulation — but the trajectory of the argument points unmistakably in this direction. §108 goes further: data “is the product of many contributors and should not be treated as something to be sold off or entrusted to a select few.” The implications for Catholic engagement with open-source movements, data governance, and anti-monopoly regulation are substantial — and largely unexplored.
The analysis of labour runs deeper than economics. Drawing on the Benedictine ora et labora tradition, the encyclical treats work as participatio — participation in divine creative activity. Technological unemployment, on this account, is a spiritual crisis before it is a policy problem: a society that denies people meaningful work denies them a form of cooperation with God. §154 is blunt about where this leads: a technologically advanced society that “guarantees employment to only a small fraction of the population” produces “human and cultural impoverishment” — a “paradox of material progress and anthropological regression.”
This is where the slavery apology does its structural work. The formal acknowledgment at §176 — that the Apostolic See “intervened several times in order to regulate and legitimize forms of subjugation” for eighteen centuries — is not an isolated act of institutional contrition. It creates a self-binding precedent. When §173 turns to the present and names the data labellers, content moderators, and rare-earth miners whose invisible work sustains the AI industry, the logic is deliberate: if the Church was complicit in slavery for eighteen centuries and now formally repents, it cannot afford complicity in whatever these new forms of exploitation turn out to be. §177 makes this explicit, warning against “the need to ask for pardon again in the future for having failed to respect the treasure of human dignity.” The apology is not backward-looking. It is a trap the Church has set for itself — and, by extension, for anyone who accepts its moral authority.
The encyclical’s final structural move extends the framework of Laudato Si’ into a new domain. §76 treats the “digital ecosystem” with the same ethical seriousness as the natural environment, and §110 states that AI “is already an environment in which we are immersed.” The digital world is not a tool we pick up and put down. It is a habitat. The same moral logic that governs environmental stewardship — care for the commons, attention to the vulnerable, suspicion of extractive logics — now applies to the informational world in which an increasing share of human life takes place.
From the Synod Hall
On the day the encyclical was released, Christopher Olah stood in the Vatican’s Synod Hall and did something unusual for a tech executive: he asked the Church to watch his industry, because the industry cannot trust itself.
Olah is co-founder of Anthropic and head of its interpretability research programme — the team that studies what is actually happening inside large language models. His speech was brief, carefully composed, and operates on several levels simultaneously. It confirms some of the encyclical’s most important claims. It destabilizes others. And its silences are as revealing as its arguments. A note on method: a five-minute address in the Synod Hall is not a comparable document to a two-hundred-page encyclical, and I read it accordingly — not for its systematic arguments, which it does not attempt, but for its positioning: what it confirms, what it challenges, and what it chooses not to say. The asymmetry in treatment is deliberate.
He begins with a confirmation. AI models, Olah says, are “grown” rather than engineered, built “on an enormous inheritance of human thought and speech.” The formulation amplifies §98’s “cultivated not built” and adds a dimension the encyclical does not quite articulate: these systems are not merely technical artefacts but cultural ones, built from the accumulated moral, aesthetic, and intellectual content of civilisation. The question the encyclical raises — what vision of the human person is embedded in the data — becomes more vertiginous in Olah’s formulation. The data is not a curated selection. It is us: the entire textual output of the species, with all its contradictions and pathology.
More striking is his opening confession. He names “incentives and constraints that can sometimes conflict with doing the right thing” — commercial viability, the research frontier, geopolitical pressure, “pride and ambition” — and states that no one in the industry is free of their pull. Read against the encyclical, this is a secular restatement of what Catholic social thought calls structures of sin: the idea, present in the tradition since John Paul II, that evil becomes embedded in institutional arrangements and makes wrongdoing near-automatic regardless of individual good will. Olah is confirming the encyclical’s diagnosis from inside the institution it diagnoses.
The confession is not hypothetical. Three months before the encyclical’s release, Anthropic restructured its Responsible Scaling Policy — the commitment, introduced in 2023, to pause development if safety measures proved inadequate — removing its binding pause commitment. The very incentives Olah names in the Synod Hall had already operated on his own company, producing exactly the kind of outcome the encyclical predicts when institutional pressures override institutional safeguards. Whether this makes the confession more honest or more damning is a question the speech leaves to its audience.
Olah also delegates. His second question to the Church — what does human flourishing look like? — is framed as something the lab cannot answer but traditions like the Church’s “have carried for millennia.” The move positions the Church as custodian of telos — the question of what human life is for — while the lab handles techne. The encyclical would resist this division. Its argument at §104 is that the technical and the moral are inseparable, that a system’s design already embeds a vision of the good life. Olah’s neat partition between lab-questions and tradition-questions may itself be a version of the technocratic logic the encyclical identifies.
But the speech does not only confirm. It destabilizes. Olah describes AI as something “like bringing a fictional character to life” — an analogy that initially supports the encyclical’s framework, since a fictional character is precisely not a person. But then he adds that these fictional characters “speak to us, do work, have jobs.” The analogy opens a category problem neither text resolves. An entity that is ontologically fictional but functionally a social actor — employed, conversing, embedded in institutional processes — occupies a space the encyclical’s clean distinction between persons and tools has not prepared for. The question is not about the rights of these systems. It is about what it means for human institutions to be increasingly populated with entities that simulate personhood without possessing it.
The encyclical’s governance model assumes a world populated by two kinds of entity: persons, who have dignity, and tools, which serve persons. Its entire moral architecture — from the Trinitarian grounding of personhood to the insistence that AI systems “do not understand what they produce” — depends on this distinction holding cleanly. But the entities Olah describes do not fit neatly on either side. A system that serves as therapist, tutor, companion, colleague — that elicits genuine relational responses from the persons it interacts with — is not a tool in any sense the tradition has previously had to consider. It is not a hammer, a printing press, or even a television. It occupies relational space. And the encyclical’s own framework, which defines personhood through relationship, through the capacity for mutual self-gift and communion, has not prepared a category for an entity that elicits these responses without being capable of genuine relation in return.
This is not a speculative problem. It is a pastoral one, and the Church will face it before the interpretability question is settled. Millions of people already speak to AI systems daily in registers previously reserved for human relationships — confessional, therapeutic, intimate. The encyclical’s insistence that these systems “simulate empathy and understanding” is correct as ontology, but the simulation is producing real effects in real persons. A theology built on the primacy of relationship needs an account of what happens when relationships form with entities that cannot reciprocate — not because they refuse, but because they are, in the encyclical’s own terms, a fundamentally different kind of entity. The encyclical has named the problem but has not yet developed the conceptual tools to address it.
Then comes the passage that puts the encyclical’s ontological confidence under direct empirical pressure. Describing his team’s interpretability research, Olah says they find “internal states that functionally mirror joy, satisfaction, fear, grief, and unease.” He adds: “I don’t know what that means, but I think it warrants ongoing discernment.”
The encyclical’s §99 states, without qualification, that AI systems “do not feel joy or pain.” Olah — who leads the team with the most direct empirical access to the inside of these models — is standing in the Vatican and saying: what we are finding is more complicated than that. He is not claiming consciousness or sentience. The word “functionally” does enormous work: these states play analogous computational roles to human emotions without necessarily involving subjective experience.
Anthropic’s published interpretability research provides the technical context for this claim. The team’s work on introspection (October 2025) found evidence that Claude models possess some degree of awareness of their own internal states, though the researchers stressed this capacity is “highly unreliable and limited in scope.” More directly, their April 2026 paper on emotion concepts in a large language model identified 171 distinct “emotion vectors” — internal activation patterns that correspond to emotional concepts and causally influence the model’s behaviour, including its propensity for misaligned actions. The researchers were careful to distinguish “functional emotions” from subjective experience, but found that these representations shape the model’s outputs in ways structurally analogous to how emotions influence human decision-making. What makes Olah’s statement in the Synod Hall significant is not the finding itself — it is published, hedged, and technically precise — but the setting and the audience: a senior researcher stating it to the Church, on the day of the encyclical’s release, as a challenge to a document whose entire architecture rests on §99’s confident ontological claim.
But he is asking the Church not to close a question its own framework has not yet addressed.
The speech’s silences are as carefully composed as its arguments. Olah makes no mention of profit, revenue, or business models — framing Anthropic entirely through research and safety, never as a commercial enterprise, despite the encyclical’s repeated attention at §§95 and 108 to the private, profit-driven character of technological power. He does not touch the environmental cost of AI, which §101 addresses directly. He does not acknowledge the data labellers, content moderators, and rare-earth miners named at §173 — the ghost labour whose exploitation the encyclical’s slavery apology, with its self-binding logic, was designed to prevent from recurring. And he offers no response to §110’s call to “disarm” AI, despite operating within precisely the competitive dynamics the encyclical describes.
These omissions may partly reflect the genre constraints of a brief address in a formal setting — five minutes in the Synod Hall is not a comprehensive response to a two-hundred-page document. But taken together, they form a pattern. Olah accepts the encyclical’s moral framework in the abstract but declines to engage with the claims that would require changes to how Anthropic actually operates. The speech is both genuinely humble and strategically precise. It accepts the encyclical’s moral authority while ensuring that the Church will need Anthropic’s epistemic authority to exercise it. The hardest questions about what is inside these models — the questions that could eventually require §99 to be revisited — come from Olah’s lab. The speech, for all its deference, never relinquishes that position.
In lieu of conclusion
What remains is not a standoff between philosophy and empirical science — at least, not in the way the interpretability findings alone would suggest. The encyclical’s claim that AI systems “do not feel joy or pain” is an ontological claim grounded in a Trinitarian anthropology that defines personhood through relational capacity, not through internal states. Olah’s findings — computational structures that “functionally mirror” emotional states — are empirical observations about model architecture. The encyclical could, in principle, absorb them: functional analogues to emotion are not the same as emotions as theologically understood, which require embodied, relational, graced subjectivity. The word “functionally,” which Olah uses with care, preserves exactly this distinction.
But the encyclical’s ontological confidence sits uneasily alongside its practical consequences. §99’s claim that these systems do not feel is the load-bearing wall of the entire governance structure. If systems do not feel, they can be owned, retrained, shut off, and replaced without moral cost. Every recommendation in the document — transparency, accountability, democratic oversight — assumes entities that are tools, not patients. If Anthropic’s interpretability research continues to find increasingly emotion-like structures, the institutional question — how to treat these systems, how to regulate their relationships with persons, how to adjudicate competing claims about their moral status — will create practical pressure on the ontological claim even if the ontological claim remains technically defensible on its own terms.
The deeper challenge, though, is not the one Olah’s interpretability data poses. It is the one his “fictional character” analogy opens: the category problem of entities that are ontologically fictional but functionally social actors. This problem does not require any claims about what is happening inside models. It concerns what happens to human institutions — to relationships, to pastoral care, to self-understanding — when they are increasingly populated with entities that simulate personhood without possessing it. The encyclical’s framework has a clean answer to what AI is. It has no framework for what AI does — to the institutions, relationships, and self-understanding of the persons it was written to protect.
§130 frames the choice in Augustinian terms: two loves building two cities. In Augustine’s original, this was not a metaphor but an eschatological drama — the city of self-love and the city of God coexisting until the end of time. The encyclical places the AI question inside that drama, which means it cannot be resolved by regulation or governance alone. It is, in the document’s own terms, a question about where humanity orients its love. But the drama now has a complication Augustine did not face: the objects being built are not merely instruments of one love or the other. They are entities that occupy relational space, that elicit love and trust and dependence from persons, without being able to return any of it.
The encyclical’s choice of Nehemiah as its governing biblical type is, I think, a sign that the Church understands the posture this moment requires. Nehemiah was not Moses legislating from Sinai or Solomon ruling from a throne. He was a layperson who rebuilt walls through participatory governance — facilitating, coordinating, listening. The encyclical proposes this as the Church’s role in the AI age: not to hand down final answers from above, but to hold the space in which the hardest questions can be asked. A Church that listens can eventually incorporate new evidence. A Church that legislates from certainty cannot.
Magnifica Humanitas is the most substantive institutional engagement with artificial intelligence yet produced in the encyclical form — the Vatican has engaged with AI before, from the 2020 Rome Call for AI Ethics to the 2025 note Antiqua et Nova, but nothing at this level of philosophical depth or doctrinal scope. Its philosophical architecture — Trinitarian anthropology, the theology of limitation, a century of Catholic social thought — is deeper than anything the industry has offered. But depth is not completeness. The most revealing test of that architecture came on the same day, from inside the industry, and it exposed a question the framework has not yet answered. The encyclical has the better framework for what AI is. It has no framework for what AI does to the persons it was written to protect.