“The Vatican is not just concerned about the theological implications of AI, but also about who holds the power to define the moral framework of these systems.”
That is a very polite way of saying the Pope is terrified of being sidelined in the biggest shift in human agency since the printing press. The Church has spent centuries as the ultimate arbiter of morality, but that monopoly is under threat from a few thousand H100s and a handful of engineers in San Francisco.
The play here is subtle. According to Wired, the Vatican isn’t just sending strongly worded letters or hosting symposiums; they are attempting to place “men inside” the labs. Specifically, they want a seat at the table at Anthropic. The goal is to ensure that “human dignity” and Catholic social teaching are baked into the alignment process. It’s a soft-power maneuver designed to influence the “Constitutional AI” that Anthropic uses to keep its models from going off the rails.
It is an odd image: the oldest bureaucracy on earth trying to influence the newest, fastest-moving one. The Vatican is effectively acting like a corporate lobbyist (though with better robes) trying to convince a high-frequency trading firm to trade “ethically.”
Do we actually believe that a few meetings with priests will stop a race to AGI?
The friction here isn’t just about religion; it’s about the fundamental difference between dogma and weights. The Vatican deals in eternal truths and immutable laws. AI labs deal in probabilistic tokens and RLHF (Reinforcement Learning from Human Feedback). One moves in centuries; the other moves in epochs.
Anthropic’s “Constitution” is already a set of norms—a list of rules that tells the model how to behave. From the Vatican’s perspective, this is the perfect entry point. If you can slip a few paragraphs of Catholic social teaching into the constitution, you’ve essentially hard-coded a specific brand of morality into the latent space of the model. (Which is a lot of bureaucracy for something that can be bypassed with a clever system prompt).
But this assumes that the labs actually care about the “moral framework” of their systems beyond avoiding lawsuits and bad PR. In reality, alignment is usually a game of minimizing toxicity and maximizing helpfulness. Adding a layer of theology to that doesn’t change the underlying math; it just adds another set of constraints for the model to navigate. It’s like trying to teach a toddler the intricacies of the Magna Carta while they’re in the middle of a temper tantrum.
The real conflict will arise when the Vatican’s definition of “human dignity” clashes with the utilitarian goals of the labs or the permissive nature of the open-web data the models are trained on. You cannot “align” a model to a specific faith when the model’s primary function is to be a general-purpose tool for a global, secular audience.
It’s a prestige play.
The most likely outcome is a symbiotic relationship based on mutual vanity. The Vatican gets to feel like it’s shaping the future of intelligence, and the labs get the prestige of saying they’ve “consulted with the Holy See” to make their AI more “ethical.” It’s a PR win for both sides that changes absolutely nothing about the actual training run.
By Q4, we will see the Vatican release a formal “AI Ethics” directive that the major labs will treat as a PR checkbox rather than a technical constraint. The engineers will nod, smile, and then go back to tweaking the temperature and top-p settings. The Pope might have the attention of the industry, but attention is not the same as control. In the world of LLMs, the only one with real power is the person holding the keys to the compute.
