Pope Leo XIV's AI Encyclical: What Builders Must Know (2026)

Workers, Labor Displacement, and Digital Colonialism

The section on labor is the most practically grounded passage in the encyclical. Pope Leo explicitly names the categories of workers being displaced by AI: writers, coders, analysts, designers, educators — knowledge workers who spent decades building expertise that large language models can now approximate at a fraction of the cost. The document calls for technology to not “lead to unemployment in the name of reducing costs and increasing profit,” and expresses hope for a renewal of labor organizations capable of representing workers in the AI economy.

But Magnifica Humanitas goes further than most job displacement discussions. It links AI’s economic model to what it calls a form of digital colonialism: the underpaid workers in the Global South who annotate training data, moderate harmful content, and label millions of images are described as the invisible foundation of the AI economy, often working in conditions the encyclical compares to modern serfdom. This is a documented reality. Studies by the Oxford Internet Institute and investigative reporting have detailed the conditions of data labelers in Kenya, India, and the Philippines who earn $1–$3 per hour making frontier AI models function.

For AI developers, the challenge is not abstract. The model you are calling via API was made possible by labor practices your organization almost certainly has not audited. Magnifica Humanitas does not provide a procurement checklist, but it does establish a moral principle: you cannot disclaim responsibility for the full supply chain of your AI system.

AI in Warfare: Accountability Must Not Collapse into the Machine

Perhaps the most urgent section of Magnifica Humanitas addresses autonomous weapons. Pope Leo XIV explicitly condemns lethal autonomous weapons systems (LAWS) — drones, battlefield robots, and AI decision systems that can select and engage targets without meaningful human control. The key phrase: “Accountability must never be collapsed into the machine.”

This is not merely a theological position. The International Committee of the Red Cross, the UN Secretary-General, and dozens of the world’s leading AI researchers have called for a binding international treaty on LAWS. Multiple states — Austria, New Zealand, Chile among them — have endorsed a ban. The Trump administration’s reported ban on federal agencies using Anthropic’s technology, after Anthropic declined to allow Claude to be used for autonomous targeting, adds a pointed contemporary dimension to this debate.

The encyclical also addresses AI’s subtler military applications: disinformation systems, mass surveillance infrastructure, and predictive targeting algorithms. All are subject to what the document calls “the most rigorous ethical constraints.” For developers working adjacent to defense or government contractors, this framing matters. The question is not just whether your system fires a weapon. It is whether it supports systems that do.

Developer Responsibility: Every Design Choice Reflects a Vision of Humanity

This is where Magnifica Humanitas speaks most directly to the AI builder community. The encyclical makes an argument that has gained significant traction in AI safety and alignment research but rarely from a moral authority of this scale: AI is not a morally neutral tool.

“AI developers bear a particular ethical and spiritual responsibility, for every design choice reflects a vision of humanity.”

The choices made during model development — what data to train on, what behaviors to reinforce with RLHF, what guardrails to implement, what values to encode in system prompts and constitutions — are moral choices, whether or not the engineer making them thinks of them that way. A training dataset that skews heavily toward English-language, Western-context documents encodes a particular vision of which voices and knowledge systems matter. An RLHF process that optimizes for helpfulness without adequate safeguards on harm encodes a particular trade-off between capability and safety.

This resonates directly with Anthropic’s own Constitutional AI approach, which attempts to make these value-encoding choices explicit and auditable rather than implicit and opaque. It also connects to the broader interpretability research agenda that Olah himself has pioneered: if you cannot understand what your model is doing internally, you cannot know what vision of humanity it is encoding.

For practicing engineers and product builders, Magnifica Humanitas’s challenge is practical: have you explicitly articulated what values your system reflects? Can you audit whether it embeds biases about which problems are worth solving, which users matter, or which behaviors are acceptable across cultural contexts? The encyclical does not provide engineering specifications. It insists that ignoring these questions is itself a design choice — and that design choice has consequences at scale.

The Transhumanism Critique: What It Does and Does Not Say

Magnifica Humanitas takes a careful but clear position on transhumanism and posthumanism — the idea that technology can and should overcome human biological limitations, and the related view that humanity is simply one possible form of intelligence among many. The Pope does not reject technological enhancement outright, but argues that the underlying philosophical impulse — the desire to escape human limitation through technology rather than fulfill it through love and community — reflects a flawed anthropology.

The encyclical proposes a “Christian humanism” in which human beings “are not confined by the boundaries of their own nature; rather, they are called to self-transcendence, not through an escape from reality or a contempt for their limitations but through their fulfillment in love.” This section will be most contested outside religious contexts. But even for secular AI researchers, it raises a legitimate concern: AI development as currently practiced often treats human beings as optimization targets — problems to be solved, preferences to be predicted, behaviors to be modified — rather than as subjects with irreducible worth and dignity that cannot be captured by any objective function.

The Anthropic Connection: Why Olah Was at the Vatican

The most striking aspect of the encyclical’s release was its co-presenter. Pope Leo XIV broke with tradition to personally oversee the public release of the 235-page document alongside Christopher Olah, Anthropic co-founder. Past popes have delegated encyclical presentations to cardinals or senior Vatican officials. The presence of one of the AI safety community’s most respected researchers signals that the Church is engaging with the technical AI community directly — not simply commenting from outside.

Olah’s research on neural network interpretability — understanding what concepts AI models actually represent and how they reason — is foundational to the project of making AI systems that humans can genuinely trust, oversee, and correct. Magnifica Humanitas calls explicitly for AI that can be audited, understood, and held accountable. The theological argument and the technical safety agenda are aligned on a core point: systems that are opaque to human understanding cannot be corrected when they go wrong, and systems that cannot be corrected will eventually cause harm at scale.

Five Practical Takeaways for AI Builders

Magnifica Humanitas is a 235-page theological document. Most AI developers will not read it in full. Here is the distilled version of what it is actually asking of the people who build AI systems:

1. Read the labor section seriously. If your AI product eliminates jobs or depends on data annotation supply chains, have you examined those supply chains? Are the workers who made your model possible — the annotators, the moderators — being compensated fairly?
2. Think about power concentration in your architecture decisions. Proprietary API lock-in, single-provider model dependencies, and closed training data all contribute to the AI monopoly dynamic the encyclical critiques. Open models, self-hosted infrastructure, and interoperability matter beyond technical preference.
3. Take interpretability seriously. “Every design choice reflects a vision of humanity” is not just a moral claim — it is a product design claim. Systems that can be audited and understood are systems that can be improved and corrected. Observable AI systems are safer AI systems.
4. Build accountability into autonomous systems. Whether or not you are building weapons, the principle applies: if your AI system takes consequential actions — denying loans, flagging content, making hiring decisions — there must be a human accountable for those actions. “The algorithm decided” is not accountability.
5. Magnifica Humanitas is not anti-AI. It is pro-humanity. These are not the same thing. Conflating a call for ethical constraints on AI development with opposition to AI development itself misreads the document — and misses the most important question it raises.

Why This Document Matters Beyond Its Source

Magnifica Humanitas will not rewrite your codebase or change your deployment pipeline. But its publication is a marker of something significant: AI has become consequential enough that one of the world’s oldest and most influential moral institutions felt compelled to make it the central subject of its first major teaching document in years. The Church reaches approximately 1.4 billion Catholics globally, with substantial institutional influence in healthcare, education, labor organizing, and policy in dozens of countries. Its framing of AI ethics will shape legislation, procurement decisions, and public discourse in ways that will affect the market in which AI products compete.

The encyclical’s central argument deserves to sit with anyone who builds AI systems: the question is not only what your technology can do, but what vision of humanity it encodes. That question does not have a checklist answer. It has to be revisited with every significant design decision, every new capability, every deployment context. Magnifica Humanitas is asking AI developers to treat that question as seriously as they treat performance benchmarks and cost optimization.

That is a reasonable ask. And the fact that it is now coming from the Vatican, in coordination with one of the leading AI safety researchers in the world, suggests that the community of people asking it is larger, and more serious, than it has ever been.