Catholic AI Brief: Changing Medium
Pangram can't grok the Pope
1. Robots Talking To Robots
Almost inevitably there were a number of articles asking "Did Pope Leo use AI to write the encyclical?" I was very skeptical of these claims as they rolled in, especially since most seemed to be completely unaware of how an encyclical is put together, treating it as a sort of "The Pope writes a college essay" scenario. In fact, the encyclical was certainly drafted by multiple people in at least two languages (English and Italian) and then consolidated into a master version in one language, with translations then made of that version.
Most of the takes in favor of this theory lean on results from the AI-use detector Pangram which claims a below 0.05% false positive rate for their tests. The problem – and the reason to take this idea with a bucket of salt – is that the encyclical writing process is so weird that it's impossible that Pangram could be properly calibrated to evaluate it:
- Many academic authors drafting different sections in different languages
- from 2000 years of Church context
- into section drafts then edited by Dicastery principals
- combined into an overall draft approved by the Pope himself
- edited and proofread with an insane level of attention
- translated into multiple languages each using a sort of organic Papal-style unique to Church documents
There is no document that has been created since the advent of generative AI that comes even close to this. And beyond that – AI-detection systems are intrinsically unstable. Anyone with a decent knowledge of AI tropes and patterns could easily write something that gets a very high score on any of these services. And that's the point – generative AI writing works in a very similar way to how encyclicals are drafted, just on a much less refined level, using a broader, shallower set of vectors.
2. AI TLDR;
There was an unprecedented volume of instant takes on the encyclical, because the knee-jerk reaction was to feed it into Claude and mine the cool quotes and a clean overview. Then came a wave of commentary warning against this, and a decent amount of smug proclamations that "I'm actually going to read it, and you shouldn't comment on it till you have as well."
But is using AI to summarize the encyclical or pull out insights so bad, really? Sure, it's a less rigorous way to engage with the document, and it does have a second order effect of pumping the internet full of sloptakes. But how many of the people doing that were really even going to read the thing? Or would they have just skimmed the secular press articles? ("Pope goes to war with AI", "Pope Leo calls for AI disarmament", "Did Pope Leo use AI to write it?")
Is it, in fact, time for the Church to realize that this is how the majority of people are using AI these days, and adapt how the message is promulgated? The current AI moment is like the advent of the printing press. We can ignore it and continue with business as usual, or we can accept that the medium is the message, and take a proactive approach to crafting that medium.
Put another way – why can't I have a conversation with the encyclical? Especially if the alternative is ignoring it?
3. Google ships the medium
That closing question isn't rhetorical anymore. Two weeks ago at I/O 2026, Sundar Pichai opened the keynote by welcoming us to "the agentic Gemini era" — Google is no longer treating conversational agents as a feature inside one product, but as the connective tissue running across all of them. Antigravity is an agent-first development platform, complete with subagents, hooks, and async task management. Search itself extends into AI Mode follow-ups, and Google is previewing Search agents that run 24/7 over the web on a user's behalf. The ground isn't shifting under the Church's feet — it has already shifted. The question is no longer "should we do this," it's now "what does ours look like?"
(Disclosure: this is what we've been building at Highland — our Unvarnished service already runs on six Church sites and is a formal A2A (Agent-to-Agent) provider.) The bet underneath that work is simple: as secular agents proliferate, Church sites that don't participate in A2A will get answered about by agents reading Wikipedia, rather than answered for by agents talking to the Church.
Thanks for reading! Reply with a take, a tool, or a correction. Back next week.
-- Sid