The Human Exception

Section One

A Date Chosen With Intent

The document is called Magnifica Humanitas — "Magnificent Humanity." It was signed on May 15, not released until May 25. That ten-day gap was time enough for the symbolism of the date to land: 135 years earlier, on May 15, 1891, Pope Leo XIII had published Rerum Novarum ("Of New Things") — the encyclical that helped reshape the moral architecture of industrial-era capitalism and gave Catholic Social Teaching its enduring spine. The choice of anniversary was not accidental. Pope Leo XIV, born Robert Francis Prevost, was signaling from the first stroke of the pen that this document belongs to a tradition — that it has lineage, weight, and precedent.

The encyclical addresses AI directly and at length: the opacity of algorithmic systems, the concentration of transformative power in the hands of a handful of private companies, the risks of disinformation and manipulation, the threat to human labor and dignity, the creeping logic of transhumanism that would optimize the human person out of existence. It calls for robust government regulation of AI and insists that its development cannot be governed by profit alone. It uses phrases like "anti-human vision" and warns against the emergence of "new forms of dehumanization."

These are not novel concerns. Researchers at DeepMind, Anthropic, and dozens of AI safety labs have articulated versions of them. The UN Secretary-General has raised them. The EU AI Act attempts to legislate around some of them. What is novel is who is saying this, in what form, and with what institutional infrastructure behind it.

The presentation at the Vatican — where Olah spoke alongside the Pope, acknowledged Anthropic's structural incentive problems, and called on religious communities and civil society to "take this seriously, to look closely, and to push events in a better direction" — was not a photo opportunity. It was a negotiation between two very different kinds of authority over a question neither can answer alone: what kind of future does artificial intelligence make?

Section Two

What the Encyclical Actually Says

Magnifica Humanitas is not a technical policy document. It does not propose regulatory thresholds, compute ceilings, or liability frameworks. What it does instead is something that turns out to be rarer and harder: it provides a moral grammar for thinking about AI at the level of first principles. For scientists and technologists accustomed to evaluating governance proposals on their empirical merits, this may initially feel like the wrong register. Stay with it.

The encyclical identifies a danger beneath all the specific dangers — a danger it calls the reduction of the human person to something to be optimized. It addresses transhumanism (the view that technology should overcome human biological limits) and posthumanism (the view that humanity may be surpassed or merged into machines) not primarily as science fiction but as philosophical tendencies already visible in how AI development is discussed by some of its most prominent advocates. When a system is described as eventually replacing human judgment wholesale — not assisting it, not augmenting it, but replacing it — the encyclical sees an implicit anthropology: an assumption that the human is just a less efficient version of what the machine will become.

On concentration of power, the encyclical is direct. AI development is happening in a handful of wealthy nations, within a handful of private institutions, with governance structures that are largely self-defined. This is not primarily a criticism of any individual company; it is a structural observation. The document calls for transparency, accountability, and human responsibility in decisions that affect human dignity — and insists these are not optional features that can be bolted on after deployment, but prerequisites for any legitimate development process.

The encyclical also raises the question of workers: who benefits from the productivity gains AI delivers, and who bears the cost of displacement. This is where the Rerum Novarum parallel sharpens most clearly. Leo XIII wrote in the age of factories. Leo XIV writes in the age of large language models. The underlying question is identical: when a technology concentrates economic power and disrupts the labor market at scale, what does justice require?

Olah's willingness to say this at the Vatican — not in a crisis post-mortem or a Senate hearing, but at a deliberate public event alongside the Pope — is itself significant. It represents a choice to subject the industry to a moral vocabulary that does not originate from within it. The Church brings a framework that pre-dates the incentive structures of venture capital and compute markets, and is accountable to communities that will never appear on a product roadmap.

Section Three

135 Years of Moral Infrastructure

To understand why this matters beyond the Catholic world, it helps to understand what an encyclical actually is and what Rerum Novarum actually did.

Rerum Novarum was published in 1891, when industrial capitalism had produced both staggering productivity and staggering suffering. Child labor was routine. Working days were unregulated. Wages were set by pure market power. The socialist answer — collective ownership of the means of production — was gaining traction. Leo XIII offered a third position: private property is legitimate, but property rights are bounded by obligations to workers and to the common good. Workers have rights — to fair wages, to organize, to conditions that preserve their dignity. Capital cannot treat labor as just another input to be minimized.

The document did not pass a law. But it gave Catholic political movements, trade unions, and social reformers across Europe and Latin America a moral vocabulary that shaped labor law for the next century. Catholic social teaching became an intellectual tradition with genuine influence — not because the Vatican can compel governments, but because a clearly argued moral framework, delivered through institutions with global reach, can shift the terrain on which political arguments are made.

This tradition matters for a secular audience not because the Catholic Church is always right — it isn't, as its own history makes abundantly clear — but because moral frameworks with global institutional backing are extremely rare in the AI governance space. The AI safety community produces papers. The EU produces regulations. Think tanks produce frameworks. None of them have 1.4 billion people embedded in their institutional network. None of them have 135 years of precedent for doing this exact kind of work.

It's also worth noting what the encyclical is not doing. It is not calling for a halt to AI development. It is not declaring AI intrinsically evil. It is not demanding that algorithmic decisions defer to clergy. The document explicitly affirms AI's potential to benefit humanity — in medicine, in scientific discovery, in addressing climate change. What it is doing is insisting that these benefits cannot be distributed by market mechanisms alone, that the humans affected by these systems have rights that precede any contract with a platform, and that the question of what AI is for must be answered by democratic communities, not by the companies building it.

Section Four

What Happens Next

The question of AI governance is currently stuck in a peculiar loop. Everyone agrees that something must be done. The technical community produces safety research that gets cited in policy discussions. Policy discussions produce frameworks that are too vague to enforce. Enforcement mechanisms run into jurisdictional problems. And in the meantime, the systems get more capable at a rate that outpaces all of the above.

What Magnifica Humanitas introduces into this loop is not a solution — it's a different kind of pressure. An encyclical doesn't write law. But it does mobilize communities. The Catholic Church has hospitals, schools, universities, and social service organizations in nearly every country on Earth. When a moral framework arrives with this kind of institutional infrastructure attached, it doesn't just shape intellectual debate. It shapes the organizations that implement AI in clinical settings, in education, in social services — the places where abstract governance questions become concrete decisions about which person gets which outcome from which algorithm.

There are real gaps in the document. It does not engage at depth with the technical architecture of modern AI systems — the attention mechanisms, the reinforcement learning from human feedback, the emergent capabilities question. It is more comfortable in the language of moral philosophy than in the language of compute scaling laws. Technologists reading it will periodically want footnotes that aren't there. These are genuine limitations.

But the absence of technical depth is, in a sense, the point. The most consequential governance questions about AI are not primarily technical. They are questions about who decides, who benefits, who bears the risk, and what kind of future is worth building. These are political and moral questions that technical expertise alone cannot answer — and the AI field has sometimes acted as though they can. A document that presses on those questions with moral seriousness, backed by genuine global reach, is filling a gap that no safety paper and no regulatory sandbox can fill.

Olah's presence at the Vatican is a signal worth parsing carefully. Anthropic has staked its identity on "responsible development of AI for the long-term benefit of humanity." That framing has a specific meaning within the technical AI safety community. What the Vatican event suggested is that "benefit of humanity" is a phrase that requires interlocutors outside that community — communities with different relationships to power, different institutional memories, and different stakes in the outcome. Inviting a religious institution into that conversation is an acknowledgment that the answers won't be found inside any single tradition, including the AI safety tradition itself.

The church drew a similar line in 1891. It took decades for that line to become law, and the process was contested and incomplete. The line drawn in Magnifica Humanitas will face the same struggle. But 135 years of moral infrastructure is not nothing. In a governance debate that is desperately short of durable, cross-cultural frameworks, it may turn out to be exactly what was missing.