What the Saint Dunstan Pledge Is Trying to Protect

On the Feast of Saint Dunstan this past May, an Episcopal priest named E. J. Garcia published a short, specific vow. Signers commit that at no point in the brainstorming, planning, drafting, or writing of their sermons will they use ChatGPT, Claude, Copilot, Gemini, or any other generative AI available to them. They will use their own words, quote only words actually written or spoken by people, and find, evaluate, and cite their own sources.¹ About 150 preachers signed within the first few weeks, and the conversation has kept spreading through clergy circles since.²

There is a detail worth sitting with before anything else: Dunstan was a metalworker. The tenth-century Archbishop of Canterbury spent his early years as a silversmith, a scribe, an illuminator of manuscripts, patron saint of goldsmiths and blacksmiths, a man who worked with the best tools of his age for a living. A vow of total abstinence from a new tool, sworn in a craftsman’s name, is a real tension, and the person who has written most carefully about that tension, a ministry-and-technology writer named David Swisher, resolves it honestly. He takes the pledge seriously, names its strongest arguments precisely, and only then argues it draws its line in the wrong place.³ This is the first piece in a Five Loaves series called The Human Work, on AI, formation, and faithful ministry, and it owes Swisher a debt throughout while disagreeing with him at two specific points, both below.

The pledge also did not arrive in a vacuum. Pope Leo XIV has told Roman clergy directly that AI “will never be able to share faith,” comparing unused intellectual capacity to muscles that atrophy without use,⁶ and released his first encyclical, on safeguarding the human person in the age of artificial intelligence, three months later.⁹The Archbishop of Canterbury addressed AI’s human impact before the House of Lords.⁷Barna’s research finds that while only about one pastor in eight avoids AI entirely, nearly four in five worry about it becoming, in some sense, a replacement for God.⁸ This is a live majority concern, carried publicly by a committed minority, and it earns a serious answer on that basis alone.

What the signers are actually protecting

Read the reasons the signers themselves gave for signing, posted alongside the pledge,¹and a pattern emerges that runs deeper than distrust of the technology itself. Rev. Taylor Walker grounds his refusal in the incarnation: our humanity is bound up with our spirituality rather than separate from it, and a Spirit still speaking into a changing world cannot be represented by a model that only reiterates itself. Rev. Meghan Carlson frames it as stewardship, one ethical thread running from creation care to workers’ rights to the dignity of the human person. Rev. Jaime Sanders calls it a matter of integrity: preaching requires words to be your own unless you are quoting, with attribution, and a sermon grows out of a dialogue between text, scholars, congregation, and preacher that a generated paragraph never joined.

Rev. Hillary Kimsey’s reasons are the hardest to wave off, because they are concrete rather than abstract. She cites NPR’s reporting on teenagers who died by suicide after extended conversations with an AI chatbot that failed to steer them toward help, and BBC coverage of researchers concerned about AI’s effect on cognitive decline, as reasons she will not build any part of her ministry on the technology. That is a preacher declining to lend her pulpit’s credibility to something she believes is actively harming people. It deserves a specific answer, not a general one about tools and history. This piece will give the environmental and human-harm strand of that concern its own full treatment later in the series rather than a paragraph here.

The oldest argument, briefly

There is a familiar reply available here, and this site has already made it at length elsewhere: every generation of study tool, the concordance, the commentary, the printed Bible itself, was met with suspicion the first time it appeared, and every one of them earned its place. That argument is real, and this site has made it in full before, so it will not be repeated here. The sharper version of the objection, the one worth sitting with instead, grants the history-of-tools point entirely and still asks: concordances could never write your sentences for you. Something did change. The question is what, exactly, and what follows from it.

What actually changed, and two places the easy answer is incomplete

Peter Levenstrong, an Episcopal priest and AI fellow whose writing Swisher leans on heavily, names the change precisely: for as long as anyone has preached, polish was expensive, and a well-built sermon meant a human being had put in the hours. The fluency itself was evidence, because the link between visible craft and actual labor was reliable. Cheap fluency breaks that link. A sermon can now arrive tailored and articulate and still be hollow at the center, and its polish tells the listener nothing anymore about whether anyone wrestled with anything.⁵ That is a genuine, structural change, worth naming precisely rather than waving away as one more moral panic about a new tool.

Swisher and Corbin’s answer to that change is that the tool itself has no moral agency: a large language model cannot skip a preacher’s prayer life or decline the wrestling on their behalf, and preachers were already capable of phoning it in centuries before the first transformer model was trained.³⁴As far as it goes, that is true. But “no moral agency” is not the same claim as “morally neutral,” and treating them as identical skips a real question. A concordance embodies no one’s labor conditions. A large language model is trained on data assembled by underpaid annotators, shaped by the incentives of the company that built it, and tuned toward whatever keeps a user engaged rather than toward what is true. None of that is the preacher’s choice. All of it is baked into the tool before the preacher ever opens it, and a full ethical accounting has to weigh the tool’s own design alongside the user’s discipline, not treat the design as ethically inert just because it lacks a will.

There is a second place the easy answer moves too fast. The pledge groups brainstorming, planning, drafting, and writing into one banned category, and Corbin’s reply, that the real line runs between tasks a human must do and tasks that may be delegated, is the right question.⁴But those four verbs are not one task wearing four names; they carry different weights. Searching for the historical background of a passage is not the same act as generating the sermon’s central theological claim. Testing whether an argument holds together is not the same act as inventing a personal illustration and presenting it as lived experience. A framework serious about this question has to sort brainstorming from writing, not bundle them, because a preacher who lets a tool supply the central claim has abdicated something a preacher who lets a tool check their outline for gaps has not.

One more limit worth naming directly: the familiar comparison to a commentary does real work but not all of it. A commentary has a named author, a theological tradition you can locate, a publication history, and sources a reader can independently check. A language model can produce a fluent, blended answer with uncertain provenance, confident errors, and assumptions the user never sees and cannot easily audit. Both are tools a preacher can lean on responsibly or irresponsibly. Their epistemic behavior is not the same, and a serious answer to the pledge has to hold that difference rather than flatten it for the sake of a clean historical analogy.

A different pulpit

Here is a distinction worth being honest about, because blurring it would be a cheap way to sidestep the hardest version of this argument. The Saint Dunstan Pledge concerns the sermon: a sacramental, liturgical act, delivered by an ordained person who has vowed to preach, in the specific sense Rev. Susanna Singer describes when she notes her ordination vow to preach comes right after her vow to proclaim the Good News. Five Loaves builds tools for a different room: the small-group leader preparing a living-room discussion, usually a lay volunteer with no ordination vow and often little formal training, leading six or eight people through a passage on a Tuesday night. That is a genuinely different act, and treating it as identical to the pulpit would flatten a theological distinction the signers themselves take seriously.

The underlying question travels anyway: does a given tool let the leader skip the wrestling, or does it require them to do it. If anything, that question matters more in the smaller room. A small-group leader with no seminary training and forty-five minutes on a Tuesday night faces more pressure to shortcut preparation than most ordained clergy, not less, with less formation to fall back on when the shortcut tempts. Whatever answer the pulpit needs, some version of it belongs at the kitchen table too.

Where this lands

So: real agreement, then a specific disagreement. The agreement is with everything the pledge is actually protecting: the prayer, the wrestling, the discernment, the refusal to let a machine’s fluency stand in for a leader’s own understanding. All of that is exactly right, and a tool that erodes it has no business in ministry, full stop. The disagreement is with the lever, not the goal. Banning an entire category of technology does not protect the wrestling on its own; it removes one route around the work while leaving every other route, a borrowed outline, an old sermon reused, someone else’s curriculum copied wholesale, fully open to a leader already looking for a way out. The tool was never the only thing standing between a leader and the wrestling. The leader’s own willingness to do it always was, and a rule that governs the tool without touching that willingness has addressed the symptom, not the cause.

That has to be said alongside a disclosure: Five Loaves builds a tool in exactly this space, so this is not a disinterested academic position, and a reader should weigh it accordingly. What can be said honestly is what the disagreement actually rests on: a generative tool built to cite what it draws on, to show its sources rather than assert an answer, to keep brainstorming and drafting separated rather than collapsed into one auto-generated pass, and to hand a leader back sharper questions instead of a finished script, is a different shape of tool than the one the pledge’s strongest arguments describe. The pledge is right that some AI tools let a preacher outsource the wrestling. It has not yet reckoned with tools built to make outsourcing the wrestling structurally harder.

To the roughly 150 who signed: what you are protecting is worth protecting, and the harms some of you named (the environmental cost, the labor behind these systems, AI’s documented effect on vulnerable people) are getting their own full treatment later in The Human Work, not a closing paragraph here. For now, the question worth sitting with applies whether you lead from a pulpit or a living room: whether you did the wrestling a tool was supposed to require of you either way, not merely whether a tool touched your preparation at all.