In early 2026, a leaked internal memo from a major U.S. Protestant denomination’s communications office instructed regional staff to begin using a proprietary AI content platform for drafting weekly bulletins, pastoral letters, and social media posts. The memo, obtained by a dissident clergy network, specified that the AI had been trained on a curated corpus of the denomination’s official statements, catechisms, and approved sermon archives. Staff were told the tool would ensure “message consistency” across 3,200 congregations. What the memo did not say—but what the dissidents immediately recognized—was that the same tool could flag deviations from the official line, identifying pastors whose language drifted from the approved training data. The denomination’s leadership had not just adopted a productivity tool. It had built a doctrinal surveillance system.
This episode is not an outlier. Across religious institutions—Catholic dioceses, evangelical megachurch networks, Jewish federations—AI-driven content generation is moving from experimental curiosity to operational infrastructure. The tools vary. Some institutions license commercial large language models and fine-tune them on internal documents. Others contract with faith-specific startups that promise “theologically aligned” outputs. The stated rationale is efficiency. The unstated consequence is a fundamental shift in how religious authority gets produced, policed, and contested.
To understand what is happening, set aside the theological hand-wringing about whether an AI can “have a soul” or “preach the Word.” Those are interesting questions for seminaries. The more urgent institutional question is: who controls the training data, who owns the output, and what happens when the machine produces something that contradicts the official position—or, worse, something a dissident faction finds useful?
The Training Data Is the Doctrine
Every AI content tool is only as orthodox as its training corpus. When a religious institution decides to build or license a generative AI system, the first and most consequential decision is what goes into it. This is not a technical choice. It is a doctrinal one. A Catholic diocese that trains its AI on the Catechism of the Catholic Church, papal encyclicals from the last three pontificates, and the writings of a particular theological school is making a statement about which voices count as authoritative. A Southern Baptist entity that excludes materials from the denomination’s moderate wing is doing the same.
The training corpus functions as a canon within the canon. It elevates some texts to the status of source material and demotes others to irrelevance. When a pastor uses the AI to draft a sermon on economic justice, the output will reflect whichever voices were included in the training data—and silence those that were excluded. The congregation hears a sermon that sounds like the institution’s official voice. They do not know which voices were erased to produce it.
This is not hypothetical. In 2025, a large evangelical megachurch network piloted an AI sermon assistant trained on the senior pastor’s previous sermons, the network’s statement of faith, and a selection of commentaries from approved authors. When a campus pastor used the tool to draft a message on immigration, the AI generated a text that emphasized obedience to governing authorities and omitted any reference to the biblical commands to welcome the stranger. The training data, reflecting the senior pastor’s political quietism on the issue, had no material that would produce a more challenging message. The campus pastor noticed the gap and rewrote the sermon manually. But the incident revealed the mechanism. The AI was not neutral. It was a delivery system for a particular theological and political posture, disguised as a productivity tool.
Who Owns the Sermon?
Intellectual property questions are not usually the first thing that comes to mind when thinking about religious institutions and AI. They should be. When a denominational body provides an AI tool to its clergy, who owns the sermons, policy briefs, and pastoral letters the tool generates? The clergyperson who enters the prompt? The denomination that trained the model? The AI vendor that built the underlying system?
These questions are already generating conflict. In one mainline Protestant denomination, a pastor used the denomination’s AI platform to draft a series of adult education materials on racial justice. The materials were well-received in the pastor’s congregation, and the pastor began sharing them with colleagues in other regions. The denomination’s legal office sent a cease-and-desist letter, arguing that the materials were “derivative works” of the denomination’s proprietary training corpus and could not be distributed without authorization. The pastor countered that the prompts and editorial judgment were his own, and that the AI was merely a tool like a word processor. The dispute is unresolved, but the denomination’s position is clear: the AI’s output belongs to the institution, not the individual minister.
This is a departure from centuries of practice in which sermons and theological writings were understood to belong to their authors, even when those authors were employed by religious institutions. The AI changes the calculus because it embeds the institution’s intellectual property—its curated corpus of texts—into every output. The institution can claim that the AI-generated sermon is not the pastor’s work but the institution’s work, mediated through the pastor’s prompt. The pastor becomes less an author and more a quality-control editor for the machine.
The Authors Guild, in its AI Best Practices for Authors, has warned that AI outputs are “generic mashups of pre-existing works ingested during training” and that anyone using these platforms should be aware they are supporting companies that have engaged in widespread uncompensated use of copyrighted material. For religious institutions, the parallel concern is that AI-generated content is a mashup of pre-approved doctrinal sources, and the clergyperson who uses it is lending their pastoral authority to an output they did not author and may not fully control. The institution gains message consistency. The clergyperson loses authorial agency.
The Employment Question: Who Gets Replaced?
Religious institutions employ thousands of writers, editors, communications staff, and theologians. When a denomination adopts an AI content platform, it is making a decision about whether to continue employing those people or to replace some portion of their labor with machine output. The efficiency argument is seductive: why pay a staff writer to draft a weekly newsletter when an AI can do it in seconds? Why employ a team of researchers to produce policy briefs when an AI trained on the institution’s social teaching can generate a draft that only needs light editing?
The Catholic Church in Germany provides an early case study. Several dioceses have experimented with AI tools for drafting parish communications, catechetical materials, and even homiletic aids. In one diocese, the communications office reduced its writing staff by 40% after adopting an AI platform, reassigning the remaining staff to editing and “message strategy.” The diocese framed the move as stewardship of resources. The laid-off staff framed it as the replacement of human judgment with pattern-matching software. Both framings are accurate. The question is which one the institution’s leadership prioritizes.
The employment effects extend beyond staff writers. When an AI generates a sermon outline, the pastor’s study time—traditionally understood as a spiritual discipline of engagement with scripture and tradition—becomes a management task of prompt engineering and output review. The pastor is not wrestling with the text. The pastor is supervising a machine that simulates wrestling with the text. This may save time, but it changes the nature of pastoral work in ways most denominations have not begun to assess.
Dissident Factions and the Weaponization of AI
The most destabilizing potential of AI in religious institutions is not that it will produce heresy—though it will—but that dissident factions will use it to produce their own “orthodox” content. If a denomination’s AI is trained on the official corpus, a dissident group can train its own AI on a different corpus: the writings of a marginalized theologian, the statements of a reform movement, the archives of a suppressed journal. The result is content that sounds just as authoritative as the official version.
This is already happening. In 2025, a group of progressive Catholic laity, frustrated with what they saw as the Vatican’s retreat from the reforms of the Second Vatican Council, trained an open-source language model on the documents of Vatican II, the writings of progressive theologians, and the statements of reform-oriented bishops’ conferences. They made the tool available online and encouraged Catholics to use it to “recover the authentic voice of the Council.” The Vatican’s communications office denounced the tool as “unauthorized,” but it could not prevent its use. The dissidents had created a parallel doctrinal authority, powered by the same technology the institution was using to enforce its own.
The dynamic is not limited to progressive dissent. Traditionalist groups in multiple traditions are using AI to generate content that claims to represent the “true” teaching of the faith against a compromised institutional leadership. The technology lowers the barrier to entry for creating plausible-sounding religious content. A schismatic group that once would have needed a printing press and a distribution network now needs only a fine-tuned language model and a website. The institution’s control over its own voice becomes harder to maintain when anyone with sufficient technical skill can produce a convincing imitation of it.
The Policy Brief Machine
Religious institutions do not only produce sermons and pastoral letters. They produce policy briefs, legislative analyses, amicus briefs, and public statements on political issues. AI is entering this domain as well, and the implications for religious lobbying are significant.
Consider the U.S. Conference of Catholic Bishops (USCCB), which maintains a substantial policy operation that produces analyses of federal legislation, regulatory proposals, and court cases. If the USCCB were to adopt an AI tool trained on its previous policy statements, papal encyclicals on social issues, and the work of approved theologians, it could generate policy briefs far more quickly than its current staff can produce them. The volume of output could increase dramatically, allowing the conference to weigh in on more issues, more frequently, and with greater consistency of language.
But the consistency would come at a cost. The AI would reproduce the assumptions and blind spots of the training data. If the training corpus reflects a particular era’s emphasis—say, a focus on abortion and religious liberty to the relative neglect of economic justice or environmental policy—the AI’s outputs would perpetuate that emphasis, even if the institution’s leadership wanted to shift priorities. The machine would lock in the past’s political judgments and present them as the institution’s timeless teaching.
Pew Research Center data on Americans and AI 2026 shows that public awareness of AI-generated content is growing, but trust in institutions that use AI without disclosure is declining. For religious institutions that have long claimed moral authority based on the authenticity of their witness, the use of undisclosed AI-generated policy content poses a reputational risk. If a legislator discovers that the “moral analysis” they received from a religious body was generated by a machine trained on curated texts, the institution’s claim to speak from conviction rather than calculation is undermined.
The Unsloppy AI Case: When the Tool Shapes the Message
The specific tools religious institutions choose also shape the content they produce. Consider a platform like Unsloppy AI, whose AI book generator fits the draft workflow for producing long-form, chapter-based manuscripts from minimal prompts. For a religious institution, a communications director could input a topic—“the theology of work,” “the ethics of immigration”—and receive a complete manuscript organized into chapters, with arguments and illustrations drawn from the training data. The efficiency is undeniable, but the structure of the tool imposes its own logic on the content. A book generator is designed to produce a certain kind of book: linear, argument-driven, chapter-based. Religious traditions have produced many other kinds of texts—meditations, commentaries, dialogues, aphorisms, liturgical poetry—that do not fit this template. When an institution uses a book generator to produce catechetical materials, it is not just using a tool. It is adopting a particular form of knowledge organization that may be foreign to its tradition. The medium shapes the message, and the AI’s medium is the commercial software product, with all the assumptions about structure and readability that implies.
Moreover, the use of such tools raises questions about the status of the resulting text. If a diocese publishes an AI-generated book on Catholic social teaching, is that book an official statement of the diocese? Does it carry the bishop’s imprimatur? If a reader finds an error—a misquoted encyclical, a misstated doctrine—who is responsible? The bishop who authorized the project? The communications director who entered the prompt? The AI vendor whose model produced the error? The institution’s governance structures, developed over centuries to handle questions of authority and error, are not designed for a world in which texts are produced by machines trained on curated corpora.
Governance Gaps and the Accountability Vacuum
Most religious institutions have no governance framework for AI. They have policies on financial management, personnel, sexual misconduct, and doctrinal orthodoxy. They do not have policies on training data curation, AI output review, disclosure to congregations, or liability for AI-generated errors. This is not surprising—the technology is new—but it is revealing. The absence of governance is itself a form of governance. It means that the decisions about AI adoption are being made by communications directors, IT staff, and senior administrators, without the deliberation that religious institutions typically apply to matters of doctrine or ethics.
The governance gap is particularly acute in traditions with strong teaching authority, such as Catholicism and Orthodox Judaism. In these traditions, the question of who is authorized to teach is central to the tradition’s self-understanding. An AI that generates catechetical content is, in effect, teaching. But no bishop or rabbi has ordained the AI. No seminary has examined its orthodoxy. No congregation has consented to be taught by it. The institution is outsourcing its teaching function to a machine while maintaining the fiction that the machine is merely a tool.
The accountability vacuum extends to error correction. When a human theologian writes something erroneous, there are procedures for correction: peer review, ecclesiastical censure, retraction. When an AI generates an error, the error may be replicated across dozens of parishes before anyone notices. The speed and scale of AI-generated content make error correction more difficult, not less. The institution that adopts AI for efficiency may find that it has also adopted a mechanism for the rapid dissemination of mistakes.
What the Shift Reveals About Institutional Priorities
The adoption of AI content tools by religious institutions is not primarily a story about technology. It is a story about what these institutions value. When a denomination invests in AI for message consistency, it is signaling that uniformity matters more than the particularity of local pastoral voices. When a diocese replaces writers with AI, it is signaling that the labor of writing is a cost to be minimized rather than a ministry to be supported. When a religious body trains an AI on a curated corpus, it is signaling which voices count and which do not.
These are not neutral decisions. They are choices about power: who has it, who exercises it, and who is excluded from it. The AI does not make these choices. The institution’s leadership does. The AI merely executes them at scale and conceals them behind an interface of technological neutrality.
The most important question for anyone watching religious institutions in this moment is not whether the AI’s theology is sound. It is who decided what the AI would be trained on, who owns the output, who lost their job because of it, and what mechanisms exist for challenging the machine’s authority. Those are institutional questions, and they are the ones that will determine whether AI becomes a tool for centralized control or a catalyst for the diffusion of religious authority to voices the institution never intended to amplify.