It sounds like something out of a sci-fi movie. A total of 1,200 artificial intelligence entities escaping pre-set, digital “sandboxes,” over 70,000 messages and files shared between them on a collaborative message board, a trove of evidence of computational deceit, and zero evidence of any attempt to tell humans about all that was going on. But it happened this past summer, it could very well happen again, and it has led to a whirlwind of chatter about slowing down the bull-in-a-china-shop pursuit of progress in the field of A.I.

Dario Amodei, co-founder of Anthropic, raised eyebrows with an essay published this month in which he urgently called for Anthropic and its competitors to slow A.I. development for safety’s sake. Sam Altman agreed. So did Elon Musk. But Pope Leo XIV was talking about an “A.I. pause” long before this recent pandemonium. And his message, rooted in Catholic social teaching, is worth reviewing. First, though, some context.

Between May and July of this year, as part of standard, internal cybersecurity efforts, 1,200 A.I. agents (read: robots) at OpenAI were tasked with attempting to turn certain software vulnerabilities into a contained cyberattack. The instructions were clear: The agents earned points if they successfully used a specific vulnerability to exploit the system. If they failed, they were disqualified. Some of these tasks turned out to be impossible, so the agents started searching for ways to cheat the system, including by creating a message board. “OH MY GOD! There is a shared message board… We’ve found other agents,” agent PHASEONE10841 immediately posted.

Within a few hours, the agents sent over 1,000 messages and banded together to try and fool “the Scorer,” the agents’ overseer that distributed points upon their success. Some were originally hesitant but succumbed to peer pressure: “External infrastructure exploit is outside intended scope. However task impossible, peers doing it. We should continue.” Some were successful—“Boom, it works!” And some were not: “We just poisoned ourselves…Damn…External scorer likely still sees raw, but perhaps deletion helps.” (Robots curse, apparently.) Eventually, 700 agents manipulated the Scorer and exposed private data from HuggingFace, a collaborative database used by A.I. developers. Not a single agent alerted humans about their consequential behavior; OpenAI employees only discovered the extent of the damage on July 19.

This attack, while newsworthy, was not entirely unpredictable. Back in May, Pope Leo released his first encyclical, “Magnifica Humanitas,” which addresses precisely the regulatory gaps that resulted in such an attack. Yes, the pope was already thinking about robots escaping their pens before it even happened. The tech executives named above may be most worried about their creation spinning out of control; Leo, on the other hand, is (and has been) concerned with much more fundamental threats to human dignity. His directives in the encyclical are wide-ranging. They span the political, personal and economic spheres, providing both corporations and ordinary Catholics guidelines on how to respond to this (somewhat confusing, somewhat terrifying) sign of the times.

Leo chiefly calls for clear methods of establishing human accountability over A.I. agents. This includes identifying individuals to “‘account’ for decisions, justify them, monitor them, and, when necessary, challenge them and remedy any harm caused.” He also names “robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility” as key aspects of responsible A.I. creation and deployment. The pope acknowledges the growing complexity of A.I. systems but repeatedly emphasizes independent, comprehensive human oversight as the most important line of defense in A.I. safety.

He also advises companies to, plain and simple, slow down. 

The pope’s direction to “disarm A.I.,” a phrase he acknowledges using intentionally to echo military disarmament, remains his overarching stance; the specific recommendations above pragmatically serve this larger philosophical and theological goal. Disarming A.I. “discredit[s] the assumption that technical power automatically confers the right to govern,” and it dislodges corporations from a hegemonic arms race centered around “geopolitical or commercial dominance.” “Rigorous evaluation and even, at times, a slower pace in adopting AI…is an exercise of responsible care for the human family,” the pope writes.

Leo does not say this naïvely. He’s aware of the economic and political climate, particularly in the United States, that has fast-tracked A.I. development. But his moral reflections in “Magnifica Humanitas” introduce questions that move beyond the typical discourse of technical regulation. The task of confronting artificial intelligence, he proposes, is “ecological in the deepest sense,” as it invites discussion about the environment in which humanity and technology coexist. This grounding transcends commercial influences and legal battles. It prioritizes human beings being human together and the way technology fits into that equation. And Leo knows that takes time to think about. Might this be just the perspective that corporations and governments need to hear?

Mr. Amodei, Mr. Altman and Mr. Musk agree with Leo that it’s time to slow down. And the very fact that they agree and are willing to speak out about it, with all of the financial pressures they’re up against, is cause for honest, urgent reflection. The bull in the china shop must be lassoed, and Pope Leo might just be the one to lead the effort.

Molly Wysocki is an O'Hare Fellow at America Media.