Christopher Olah sat among cardinals, theologians, and Vatican diplomats on May 25, 2026, listening to Pope Leo XIV read from a 42,300-word document titled Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence. The Anthropic co-founder was the only technology executive invited to speak at the presentation of the first papal encyclical ever devoted to a single technology.
The Pope’s core argument: artificial intelligence “demands to be disarmed, freed from logics that turn it into an instrument of domination, exclusion and death.”
That is not language from a policy paper or a congressional hearing. It is the voice of an institution that claims 1.4 billion adherents, framing AI not as a governance challenge but as a moral crisis comparable to the one that produced Catholic social teaching on labor rights 135 years ago.
What the Encyclical Actually Says
Magnifica Humanitas is organized into five chapters spanning 235 pages. Pope Leo XIV signed it on May 15, 2026, exactly 135 years after his namesake Leo XIII signed Rerum Novarum, the 1891 encyclical that defined the Church’s position on workers’ rights during the first Industrial Revolution. The timing was deliberate.
The document identifies five areas where AI threatens what Leo calls “the grandeur of humanity.”
Lethal autonomous weapons. The Pope declares the Catholic “just war” theory “now outdated” and states it is “not permissible” to entrust irreversible lethal decisions to AI systems. Machines cannot bear moral responsibility, and human conscience “cannot be replicated through computation.”
Labor displacement. The pursuit of greater profits “cannot justify choices that systematically sacrifice jobs, because the human person is an end, not a means, and the economic order must remain subordinate to human dignity.” This echoes the exact framing that Rerum Novarum used against factory owners in 1891.
Power concentration. Control of AI must not remain “in the hands of only a few.” The encyclical proposes treating data as shared goods rather than private commodities, a position that would put it at odds with every major AI lab’s business model.
Environmental cost. Large language models demand enormous energy and water resources. The infrastructure buildout increases carbon emissions at a time when nations are struggling to meet existing climate commitments.
Algorithmic bias and deception. AI responses appear objective but reflect their designers’ cultural assumptions. The simulation of empathy creates what Leo calls “false relationships,” particularly dangerous for vulnerable populations including children.
The Pope’s remedy is not a tech ban. He calls for “robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility.” The emphasis on independent oversight is pointed: it implies that the AI industry’s preferred model of self-regulation is insufficient.
Why Anthropic Was in the Room
The Vatican did not choose Anthropic at random. An April 29 delegation including representatives from Meta, Google, and Amazon met with Pope Leo XIV and held a longer discussion at the French embassy to the Holy See. But for the public presentation of the encyclical itself, only Olah was invited to speak.
Brian Boyd of the Future of Life Institute interpreted the invitation as recognizing Anthropic’s role in AI safety research, comparing it to “a papal audience with a head of state.” That framing carries weight. Anthropic has built its brand on safety, from its Constitutional AI training methodology to its public advocacy for government testing of frontier models.
Olah used his time at the podium to do something unusual for a technology executive: he asked for more criticism. He acknowledged that AI labs operate under pressures (commercial viability, geopolitical competition, raw ambition) that can conflict with doing the right thing. He called on religious communities, civil society, scholars, and governments to “take this seriously, to look closely, and to push events in a better direction.”
He also raised three questions that went beyond the encyclical’s framing. First, how to ensure AI’s economic gains reach beyond the wealthy nations where development concentrates. Second, what moral frameworks should govern how humans, families, and societies adapt to coexistence with AI. Third, and most provocatively, he described “mysterious internal structures” in AI models, including evidence of “introspection and emotional-like states” that he believes warrant ongoing examination.
That third point drew immediate attention. Olah’s reference to AI systems showing emotion-like states, delivered in a Vatican auditorium surrounded by Catholic theologians, was widely interpreted as a deliberate provocation designed to expand the conversation beyond regulation into questions about the nature of intelligence itself.
The Rerum Novarum Parallel Is More Than Symbolic
When Pope Leo XIII published Rerum Novarum in 1891, factories were displacing craft workers, wealth was concentrating in the hands of industrial barons, and governments were struggling to regulate technologies they did not fully understand. The Church’s intervention helped establish the intellectual framework for labor unions, minimum wage laws, and workplace safety regulations that took decades to implement.
Leo XIV is attempting the same move. The encyclical frames AI not as a technology problem but as an “anthropological” one, touching what it means to be human. “The convergence of automation, robotics and AI is rapidly transforming the very structure of work,” Leo writes, warning that development becomes inhumane “when it increases consumption for some while shifting costs and burdens onto others.”
The practical question is whether the tech industry will treat this document the way industrial barons treated Rerum Novarum in 1891: as a nuisance from a medieval institution. Early signals are mixed. Paolo Carozza, who chairs Meta’s Oversight Board and teaches at Notre Dame Law School, called it “a defining document for our era.” Taylor Black, a Microsoft AI executive who also directs Catholic University of America’s AI institute, said it would force industry leaders to ask, “What does it mean to be human?”
Those are supportive responses, but from people who straddle religious and corporate worlds. The labs themselves have been quieter. OpenAI, Google, and Meta have not issued public statements.
What the Industry Should Take From This
The Vatican’s moral authority is not regulatory authority. Magnifica Humanitas cannot compel OpenAI to slow down or force Google to restructure its data practices. But the document does three things that the secular AI safety movement has struggled to accomplish.
First, it frames AI risk in language that resonates outside Silicon Valley. Phrases like “the human person is an end, not a means” carry weight in legislative chambers, international forums, and communities of faith where technical arguments about alignment fall flat. The 98 state-level AI bills introduced across 34 U.S. states in 2026 were drafted by legislators who respond to moral arguments, not technical ones.
Second, it creates a permanent reference point. Encyclicals do not expire. Rerum Novarum is still cited in Catholic social teaching 135 years later. When future debates arise about autonomous weapons, algorithmic hiring, or AI-driven surveillance, proponents of regulation will cite Magnifica Humanitas the way labor advocates cite Rerum Novarum. As Simon Willison observed, the document contains “some of the clearest writing I’ve seen on the ethics of integrating AI into modern society.”
Third, it validates external oversight at a moment when the AI industry is consolidating power. Anthropic and OpenAI are both approaching trillion-dollar valuations. They are launching enterprise consulting arms, building hardware partnerships, and integrating into national security infrastructure. The Pope’s argument that “it is not enough to invoke ethics in the abstract” is directed squarely at an industry that publishes safety research while racing to ship products.
The Practitioner’s Read
From where I sit (20 years in IT operations, now implementing enterprise AI across a telecom’s infrastructure), the encyclical’s strongest sections are the ones about accountability. Section 105 argues that responsibility must extend from designers to users, but acknowledges that internal opacity makes assigning accountability for errors genuinely difficult. That is the exact problem every enterprise deploying AI agents faces today.
The document’s weakest section is its treatment of open-source AI. Leo frames the concentration of AI development in a few companies as inherently dangerous but does not engage with the argument that open models (from Meta’s Llama to GLM-5.1) distribute capability more widely than any regulation can.
Still, for a 42,300-word document written by theologians, Magnifica Humanitas demonstrates a remarkably detailed understanding of how AI systems actually work. Section 98 describes AI models as more “cultivated” than “built,” noting that developers create frameworks where intelligence grows rather than engineering every detail. That is a more accurate description of modern LLM training than most congressional testimonies manage.
The AI industry will not stop building because the Pope asked it to. But 1.4 billion people just received a detailed, theologically grounded argument for why the current pace and direction of AI development is morally unacceptable. That argument will outlast every quarterly earnings call and product launch this year.