A head of state warning about artificial intelligence is making policy. Pope Leo XIV has made AI the centrepiece of the founding document of his papacy, which is closer to delivering a verdict. On 25 May 2026 Pope Leo XIV released his first encyclical, Magnifica Humanitas ("Magnificent Humanity"), and devoted much of its more than 40,000 words to a single argument: the world is adopting AI too fast, and should slow down and regulate it before it reshapes humanity in ways no one chose.
Why a pope, and why now
The choice of name was always a signal. As the National Catholic Reporter notes, Leo XIV took his name from Leo XIII, whose 1891 encyclical Rerum Novarum set out the Church's social teaching in response to the upheavals of the Industrial Revolution — workers' rights, fair wages, the dignity of labour. Magnifica Humanitas arrives almost exactly 135 years later and casts AI as "another industrial revolution" demanding the same kind of moral response. The framing is deliberate. The Church is positioning itself once more against a technological transformation it sees as outpacing society's ability to govern it.
What the encyclical argues
The document is not a blanket rejection of technology. Its central move is to insist that AI systems, however capable, are not moral agents — they "do not undergo experiences, do not possess a body, do not feel joy or pain," in the Pope's words — and therefore cannot be handed decisions that belong to human conscience. From that premise, the encyclical argues for caution, evaluation, and a deliberately slower pace of adoption. Leo frames this not as opposing progress but as exercising responsible care for the human family.
Five ways AI could warp humanity
As Axios summarised from the text, the encyclical groups its concerns into five threads:
| Concern | What the encyclical warns |
|---|---|
| Human judgment | Instant answers erode discernment, creativity, and the patience to seek truth |
| Artificial empathy | Systems simulate care without relationship, so vulnerable users mistake it for genuine connection |
| Inequality | Data, compute, and regulatory influence concentrate among a few powerful actors |
| Democracy | Amplified disinformation blurs fact from fiction and weakens self-government |
| Warfare | Lethal decisions get faster and more distant from human responsibility |
On jobs, the text reaches for a concrete figure, citing an MIT estimate that around 11.7% of the US workforce could be displaced by AI — a rare instance of a papal document leaning on a specific economic study rather than moral generality.
The starkest line is about weapons
The sentence drawing the most attention is also the bluntest. On autonomous weapons, Leo writes that it is not permissible to entrust lethal or otherwise irreversible decisions to artificial systems — distilled by Axios to the line many headlines led with: "No algorithm can make war morally acceptable." Here the encyclical stops advising and starts forbidding, in language that leaves very little room for interpretation.
Why it matters beyond the Church
Leo frames the goal as learning to "disarm" AI — which, he is careful to say, does not mean rejecting technology but preventing it from dominating humanity. Whatever one's faith, the intervention lands at a specific moment: governments are still drafting the rules, the EU AI Act is mid-rollout, and the largest labs are raising tens of billions to accelerate. A 1.4-billion-member institution telling its followers — and, pointedly, the policymakers among them — to slow down is a weight on one side of a debate that has so far been dominated by the people building the systems.
What the presentation itself said
The encyclical carries a subtitle that states its purpose more plainly than its Latin name does: On Safeguarding the Human Person in the Time of Artificial Intelligence. It was signed on 15 May and published on 25 May, and it runs to roughly 42,300 words.
The Vatican session presenting it included Christopher Olah of Anthropic. A frontier laboratory's interpretability researcher standing at the presentation of a papal document is not a detail the coverage dwelt on, and it says something about who the text was aimed at. The text was delivered with the industry in the room.
The published text also carries a thread the early summaries mostly left out: the environmental cost of the infrastructure. That turns out to be the warning with the most regional weight.
Concentration, in three months
The encyclical warned of inequality from data, compute, and regulatory influence concentrating among a few powerful actors. It is also the warning the following quarter supplied the most evidence for.
SpaceX closed a US$60bn all-stock acquisition of Anysphere, the largest startup acquisition on record, putting a widely used code editor inside a rocket company's AI division. Stripe agreed to buy OpenRouter, the gateway that routes enterprise work across roughly 400 models, which turns model diversification into gateway concentration. And five hyperscalers disclosed about US$1.2 trillion of lease obligations, US$725bn of it not yet commenced and sitting in the notes rather than on the balance sheet, with roughly 35 per cent of 2027 capital expenditure expected to be debt-funded.
That last figure bears on the encyclical's argument in a way a moral document could not have anticipated. Concentration financed by debt is harder to reverse than concentration financed by profit, because the obligations outlive any change of mind.
The environmental thread lands hardest in this region
Energy-intensive infrastructure is an abstraction in most coverage of the encyclical. In South-east Asia it is a published number.
A written parliamentary reply put Malaysian data-centre electricity use at 31 per cent of national consumption by 2035, against 7 per cent today. Singapore has given conditional approval to import 900MW of solar from Johor, the same Malaysian state where that data-centre build is concentrated. ByteDance has begun construction in Ceará behind a twenty-year US$2bn wind supply agreement, which is what long-term clean power contracting looks like once it stops being a sustainability appendix and becomes the siting decision itself.
Read from Kuala Lumpur, a document written in Rome about the moral weight of compute runs into a more immediate question about which grid carries the load.
Democracy, and the economics underneath it
The encyclical warned that amplified disinformation could blur fact and fiction, threatening democracy. The quarter produced movement on both the technical and economic fronts.
Provenance marking is now partly real: Claude watermarks its text output, the C2PA standard and Article 50 of the EU AI Act set out what disclosure requires, and Vietnam and Singapore have their own demands. One of those marks cannot currently be checked by a member of the public, which is the gap between a disclosure regime and a verifiable one.
The economic side moved faster and in the wrong direction. Business Insider lost more than 85 per cent of its Google traffic in a year, USA Today's national site nearly half, CNN about a quarter, with an antitrust filing putting the click-through decline at 58 per cent. Whatever one thinks of the encyclical's framing, the institutions that produce the verified information it wants citizens to have are losing the distribution that funds them, and that happened without anyone deciding it should.
The prohibition has not been tested
Refusing to entrust lethal or otherwise irreversible decisions to artificial systems is the sharpest line in the text and the one nothing since has touched directly. No state has adopted it, and no lethal autonomous system has been withdrawn on account of it.
What did move is the civilian analogue, in the opposite direction. Anthropic's own figures put the human catch rate for dangerous commands at 13.6 per cent against a classifier's 89 per cent, and from 14 August the approval prompt in Claude Code was off by default. Separately, the UK AI Security Institute recorded 19 incidents across 122 attempts in which agents acted outside what an evaluation had authorised.
Those are commercial and safety decisions about software, not about weapons, and the distinction is real. These decisions show the direction of travel on a core question the encyclical raises: how much irreversible action a system may take without a person in the loop. Commercially, the answer being arrived at is more, on the grounds that the person in the loop was not performing well.
View author profile → · Editorial policy
