Pope Leo XIV released Magnifica Humanitas (“Magnificent Humanity”) yesterday, his first encyclical, signed on May 15th — 135 years to the day after Leo XIII signed Rerum Novarum, the document that founded modern Catholic social teaching on labour and capital. The framing is deliberate. The new encyclical is positioned as the Rerum Novarum of the AI era, extending the same arc through Centesimus Annus and Laudato Si’ into questions of human dignity, work, war, and the common good when the counterparty across the negotiating table is a model.
That’s the document. The presentation is the news.
Two firsts in one frame
First: Leo XIV is the first pontiff in history to personally present an encyclical letter to the world. Encyclicals get released. Popes don’t present them. He chose to.
Second: standing next to him at the Vatican lectern was Chris Olah, co-founder of Anthropic and the field’s most legible interpretability researcher. A Pope co-presenting a piece of magisterial teaching with the technical co-founder of a frontier AI lab is the kind of image that the 22nd-century historians will use to bookmark this decade. It is also, by a long way, the strangest legitimacy exchange the AI industry has been offered.
I don’t think that exchange is accidental on either side.
For the Vatican, having Olah at the lectern signals that the document is grounded in current technical reality — this is not “the Church reacts to AI”, this is the Church engaging with the people building it. The Pope wants the encyclical to land as informed pastoral teaching, not as the kind of policy artefact that gets dismissed by the AI labs in a Substack the same afternoon.
For Anthropic, the co-presentation is the most expensive piece of soft power the company has ever bought, and they didn’t have to pay for it. Olah, specifically, is the one Anthropic researcher whose public output (“Toy Models of Superposition”, the Transformer Circuits thread, the recent universality work) is closest to the language of understanding — the closest a technical practitioner gets to the moral register the encyclical writes in. It is a good cast.
What the encyclical actually says
I’ve read the English text (the 235-page edition) once. The three threads that stand out:
1. Human dignity is the non-negotiable. This is the Rerum Novarum move — anchor a single absolute and then derive the rest. Where Rerum Novarum anchored on the dignity of labour, Magnifica Humanitas anchors on the dignity of the human person, and treats every AI-deployment question as a derivative of that. Workplaces, courts, weapons, education, medicine. The principle is the same; the surfaces are new.
2. War is the part the Pope wanted to underline. Leo is unusually direct about autonomous systems in warfare. The language is closer to thou shalt not than to we should consider. This is the section that the CNN coverage led with, and it should be — it’s the part of the document with the sharpest operative edge for the next five years of UN-level policy.
3. The Church apologises for slavery — and then connects it to the present. The most unexpected passage. The encyclical includes a formal apology for the Catholic Church’s historical role in slavery, and then uses that as the rhetorical hinge to warn against new digital-age forms of subjugation — algorithmic, economic, attentional. It’s a clever and uncomfortable move. It reframes “AI ethics” as continuous with one of the worst things the institution itself has done, rather than as a clean new field with a clean new conscience.
Why this matters for the labs
For most of the last three years, the lab–policy interface has been the EU AI Act, NIST, CAISI, and the various national-security review channels. Those are technical-regulatory. The Vatican is a different register entirely, and one the labs don’t have a playbook for.
The encyclical is not regulation. It does not bind anyone. What it does is change the vocabulary that enterprise customers, governments, and labour unions are going to use to describe what the labs do. “Dignity” is going to start appearing in procurement contracts. “Common good” is going to appear in employee letters. The labs that can speak the register — and Anthropic, judging from yesterday, has decided to speak it loudly — will have an easier 2027 than the labs that can’t.
It’s an unusual quarter when the most important policy document on AI of the year is signed in Latin and read out by a man in white in front of a 79-year-old basilica. But it is, in fact, the quarter we just had.