In February 2026, Pope Leo XIV told the priests of Rome to resist the temptation to prepare homilies with artificial intelligence.
His objection was not that a machine could never arrange a competent argument or produce polished prose. He began with intellectual exercise: intelligence, like a muscle, weakens when it is not used. But he then drew a more consequential boundary.
A true homily, he said, shares faith. It emerges from the priest’s experience of knowing the Gospel and from knowledge of the community being addressed. Artificial intelligence cannot share a faith it does not possess. [1]
This was not a disciplinary decree. It was an answer delivered during a dialogue with diocesan clergy and later published in the Holy See’s official record. No penalty was announced. No congregation split. No rival ecclesial body formed around the right to preach machine-written sermons.
Yet the exchange documented something more durable than another argument about whether artificial intelligence is good or bad. It identified a point at which technical assistance ceases to be merely technical.
For Leo, the intellectual and spiritual work of preparing a homily is not incidental labor that may be removed while leaving the religious act unchanged. It is part of what makes the homily a homily.
Other religious institutions are drawing comparable boundaries around fatwas, ritual obligations, moral decisions and pastoral work. They are not drawing them in the same place. Nor are they using the same forms of authority.
A papal answer, a formal religious ruling, an administrative workflow, a committee-approved legal opinion and a survey of working pastors are not equivalent documents. Their differences reveal how religious authority responds as a machine moves from retrieving information to drafting language, from drafting language to shaping judgment, and from shaping judgment to appearing to speak for a tradition.
The documentary sample examined here is Abrahamic: Roman Catholic, U.S. Protestant, Sunni institutional and Conservative Jewish. It does not establish that comparable boundary formation is absent from Hindu, Buddhist, Sikh, Jain or other traditions; those cases remain outside the verified source set examined here.
The documented phenomenon is therefore not yet a class of AI-induced schisms. It is the stage before one: the production of rules, distinctions and institutional practices from which future orthodoxy—and perhaps future schism—may emerge.
The governing question is deceptively simple:
At what point does assistance become authorship, and authorship become religious authority?
The boundary before the break
Public discussion often treats religious use of artificial intelligence as a binary choice. A community either accepts AI or rejects it.
The documentary record is more granular.
A machine may retrieve a text, search a corpus, summarize commentary, translate a passage, suggest an outline, produce a draft, revise an argument, deliver the resulting words, answer a religious question or issue what appears to be a judgment.
Each step increases the machine’s participation. But the significance of that participation depends on the act involved.
Religious institutions are not merely asking whether an AI system is accurate. They are asking whether a particular act derives its religious character from its informational result or from the accountable human performance that produces it.
If a correct answer is all that matters, a machine may be evaluated as an instrument of retrieval and synthesis.
But some religious acts depend on more than the words or result. The identity, intention, obligation, experience or office of the actor may be part of the act itself. In those cases, an identical sequence of words may not be an equivalent performance.
That distinction appears throughout the cases that follow.
The pulpit: preparation or substitution
The Vatican record supplies the clearest personal formulation.
Leo XIV did not limit his objection to an AI delivering a homily in place of a priest. He warned against using AI to prepare one. His reasoning joined two claims that are often separated in secular discussions of automation.
The first concerned capacity. A priest who delegates the intellectual work risks allowing his own intelligence to deteriorate.
The second concerned participation. The congregation does not encounter only a text. It encounters the priest’s faith, experience and knowledge of the particular community.
On this account, the work of preparation forms the preacher before it informs the sermon. Removing that work may change the act even when the final prose remains plausible. [1]
This position exists alongside much less categorical practice among Protestant pastors in the United States.
A Barna Group survey of 442 U.S. Protestant pastors, conducted in December 2025, found that 87 percent reported using AI in some capacity. Half used it for brainstorming or idea generation, 36 percent for research into biblical or theological subjects, and 24 percent reported using it to write or edit sermons.
The same surveyed population expressed substantial caution: 71 percent described themselves as cautious about AI, while 63 percent were concerned about AI replacing the role of pastors or spiritual leaders. [9]
These figures do not establish Protestant doctrine. They do not show that a quarter of sermons were independently composed by machines, and they do not reveal how extensively each pastor revised, checked or incorporated generated material.
They document something narrower: a professional boundary that remains unsettled in practice.
The distinction between preparation and authorship was already visible in Barna’s January 2024 survey of 278 U.S. Protestant senior pastors. Forty-three percent were comfortable with AI assisting in sermon preparation and research, while 12 percent were comfortable with AI writing sermons.
The later survey measured reported use rather than comfort, so the two sets of percentages cannot be treated as a clean time series. Both nevertheless place research assistance and sermon authorship on different sides of a perceived boundary. [8]
The difference is not simply whether AI may be used. It is which part of the preacher’s work may be delegated without changing what preaching is.
The Vatican and Barna records do not represent two formal doctrinal camps. One is an official record of papal pastoral teaching. The other measures the reported practices and attitudes of a diverse sample of Protestant pastors.
Their comparison is useful precisely because the authority instruments differ. One articulates a boundary from personal ecclesial office. The other reveals a boundary emerging without a single central authority empowered to declare it.
Fatwa: the same ceiling, different lower rungs
The contrast between Egypt’s Dar al-Ifta and Dubai’s Islamic Affairs and Charitable Activities Department is more institutionally precise.
A fatwa is a formal religious-legal opinion issued by a qualified Islamic scholar, known as a mufti. It is not simply an answer retrieved from a body of texts. Its authority depends on the scholar’s training, method and understanding of the circumstances in which the question is asked.
In December 2025, Egypt’s Dar al-Ifta published a formal ruling on whether an individual may use AI applications to obtain fatwas. It began by stating that the use of AI technology is permissible in principle. The prohibition concerned reliance on current AI applications as sources of fatwa.
The ruling argued that such systems lack the scholarly formation, methodological grounding and contextual judgment required of a mufti.
Fatwas may depend on changes in time, place, circumstances, customs and the particular situation of the questioner. Current applications may produce answers that are correct, incorrect or contradictory to established rulings without possessing the human flexibility required to evaluate those conditions.
Dar al-Ifta therefore ruled that it is impermissible to rely on AI applications to obtain a fatwa and that present systems remain outside the proper scope of fatwa. [3]
Dubai has adopted a different operational model without surrendering final religious judgment to a machine.
IACAD’s use of artificial intelligence in fatwa delivery predates the present generative-AI cycle. In October 2019, it launched a public-facing Virtual Ifta service. The workflow described in 2026 therefore forms part of a longer institutional approach to machine-assisted fatwa delivery. [4]
In March 2026, IACAD described a closed internal AI system applied to the department’s previous fatwas and approved jurisprudential sources.
The system may produce answers or draft fatwas for frequently asked questions. Trained muftis then review those outputs for scholarly and jurisprudential soundness. Specialist supervisors conduct another review, while complex matters are referred for collective study by human muftis. [5]
The two institutions are not issuing contradictory answers to an identical question.
Dar al-Ifta addresses a member of the public relying on general AI applications to obtain a religious ruling. IACAD describes an internal institutional system operating on controlled sources and embedded in several stages of qualified human review.
The first rejects AI as an independent public source of fatwa. The second permits AI to participate in retrieval and drafting inside an accredited human institution.
The difference is not whether AI may be used. It is whether the machine answers the believer directly or works behind qualified human authorities.
Their shared ceiling is clear: the machine does not become the final religious authority.
Their lower boundaries differ. Dar al-Ifta excludes current public applications from the domain of fatwa. IACAD allows extensive supervised participation beneath the level of final judgment.
This is convergence on final human authority combined with divergence over permissible delegation. It is the assistance–authorship–authority gradient made institutional.
The forms of authority also matter.
Dar al-Ifta answers the new technology through a fatwa: a ruling that identifies the question, develops a legal and methodological argument, and states a conclusion.
IACAD’s document is not itself a fatwa on the permissibility of artificial intelligence. It is an official description of an administrative workflow, quoting the director of its Fatwa Department and explaining how the institution organizes production and review.
One institution declares a boundary through legal reasoning. The other builds it into procedure.
Ritual: intermediate assistance as doctrine
A 2019 Conservative Jewish responsum provides the clearest explicit formulation of the gradient.
A responsum is a written answer to a question of Jewish law. Rabbi Daniel Nevins’s Halakhic Responses to Artificial Intelligence and Autonomous Machines was approved by the Committee on Jewish Law and Standards on June 19, 2019, by a vote of 18–0–0.
The responsum considers machine agency, responsibility, life-and-death decisions and whether an artificial agent might perform a mitzvah—a religious commandment—or otherwise participate in Jewish ritual life. [6]
Its discussion of ritual assistance distinguishes between an intermediate task and the sacred act itself.
A robot or smart appliance may retrieve or position a ritual object for someone who needs it. It may carry an object or prepare it for ritual use. But the ritual must be performed by the person for whom it is an obligation.
Nevins applies the same principle to social commandments such as visiting the ill, comforting mourners and supporting the poor: a machine may assist, but a human must initiate and complete the act.
The responsum’s concluding rulings similarly state that artificial agents may facilitate mitzvot but do not thereby become human participants in the covenant. [6]
Here the analytical ladder is not merely inferred from institutional behavior. It is articulated in the legal reasoning itself.
The machine may participate in the causal chain. It may reduce physical difficulty. It may convey the required object.
What it cannot do is become the person religiously responsible for the act.
The responsum’s institutional provenance requires qualification. The Rabbinical Assembly describes approved committee responsa as official positions of the Conservative movement, but it also states that local rabbis remain the religious decision-makers for their communities and may determine practice locally.
The instrument therefore carries recognized authority without functioning as a universally self-executing command. [7]
That structure differs from both papal counsel and the Islamic cases. It nevertheless resolves the same underlying problem: a machine can be inside an act without becoming the religious actor responsible for it.
What the instruments reveal
The traditions examined here do not simply place their boundaries at different points. They exercise authority through different instruments.
The Vatican record preserves an answer by a pope speaking directly to priests. Its force derives from office, theological reasoning and pastoral address, but it is neither a canonical prohibition nor a disciplinary decree.
Dar al-Ifta uses a formal fatwa to define who may be relied upon for religious judgment and why current public AI systems fail the required conditions.
IACAD uses administrative architecture: controlled sources, trained reviewers, supervisory review and collective escalation for difficult questions. Its boundary is encoded in workflow.
The Conservative responsum reasons through legal analogy, ritual obligation and the status of non-human agents. Its authority comes through committee consideration and approval, while leaving room for local rabbinic decision.
Barna’s surveys exercise no religious authority at all. They reveal how authority is being negotiated in a decentralized professional population through individual practice, caution and informal norms.
The instrument chosen does not reveal the complete nature of authority within an entire tradition. It shows how authority is exercised in a particular dispute.
That distinction prevents false equivalence while making comparison possible.
A survey cannot be placed beside a fatwa as though both legislate doctrine. A workflow cannot be treated as though it proves theological consensus. A papal answer cannot automatically be converted into a canonical rule.
But each document records a boundary being drawn.
The labor inside the act
The homily case exposes a question that also runs beneath the other examples: when is human labor merely instrumental, and when is it part of the act’s meaning?
Leo XIV has situated his broader concern with artificial intelligence within the Catholic social tradition of work. In May 2025, explaining his choice of papal name, he invoked Leo XIII’s Rerum Novarum, written in response to the social conditions of the first industrial revolution.
He described artificial intelligence as part of another industrial transformation posing new challenges to human dignity, justice and labor. [2]
Read in that context—and this is the editorial’s reading, not an argument Leo made in the February dialogue—the warning about AI-prepared homilies also raises a labor question.
If the priest’s intellectual and spiritual preparation is part of the religious act, then automating that work is not equivalent to replacing a neutral production step.
The preparation forms the accountable speaker. The effort is not simply a cost attached to the product; it helps create the product’s religious meaning.
The same pattern appears elsewhere.
The mufti’s work is not reduced to selecting a plausible answer. It involves knowledge of method, source, circumstance, custom and the person asking the question.
The obligated Jew’s work is not reduced to ensuring that an object reaches the correct location. The human initiation and completion of the commandment matter.
The pastoral boundary suggested by the Barna surveys similarly separates administrative and research work from the relational and spiritual functions pastors are more reluctant to delegate.
In each case, artificial intelligence forces an institution to identify which human contribution may be removed without changing the act—and which cannot.
This is why the debate is not adequately described as resistance to automation. These institutions accept machine assistance at various levels.
The contested issue is whether the machine remains a tool inside an accountable human act or begins to occupy the position from which the tradition is interpreted, performed or spoken.
What would count as a schism
None of the cases examined here documents a schism caused by artificial intelligence.
Disagreement is not enough. Anxiety is not enough. A controversial service, generated sermon or experimental interface is not enough. Even contradictory rulings would not by themselves establish a schism.
A documented AI-triggered schism would require an observable institutional or communal rupture: named parties, incompatible authority claims, sustained alignment around the dispute and some consequential break in membership, governance, recognition, discipline or communal practice.
A rival body might form. Clergy might be removed or congregants depart. One authority might recognize an AI-mediated act that another declares incapable of religious validity.
Claims concerning AI-agent religious movements, autonomous doctrinal factions or machine-led expulsions should not be used to fill this category where autonomous authorship, community membership and disciplinary action cannot be established reliably.
The absence of a demonstrated schism does not make the present disputes trivial. It identifies their historical position more accurately.
Institutions and practitioners are now producing distinctions that a later conflict could harden into opposing orthodoxies. They are deciding—or, in decentralized settings, negotiating—whether AI may search but not judge, draft but not authorize, prepare but not preach, carry but not perform, advise but not bear responsibility.
Taken together, these records are evidence of doctrine in formation, although the Barna surveys document practice and attitudes rather than doctrinal rulings.
Before the schism
Religious institutions rarely need to define every boundary in advance. A boundary becomes explicit when a new practice crosses it, or when participants discover that they disagree about where it lies.
Artificial intelligence accelerates that process because it can participate in nearly every visible stage of religious production.
It can retrieve scripture, summarize jurisprudence, imitate devotional language, generate pastoral counsel, draft sermons, formulate ritual instructions and converse in the voice of an authority.
The closer its output moves to the final religious act, the harder it becomes to describe the system as a neutral instrument.
Across the Abrahamic cases examined here, the permitted role of the machine ranges from research and drafting to administrative and ritual assistance. Final judgment, sacred obligation and accountable performance remain assigned to a human actor.
Who selected the sources? Who framed the question? Who drafted the language? Who checked the reasoning? Who delivered the answer? Who possesses the obligation, experience or office from which the act derives authority? Who can be held responsible when it is wrong?
Does the act derive its religious character from an informational result, or from the accountable human performance through which that result is produced?
Those questions precede schism. They are also what make one possible.
The first durable AI-related religious split may not begin with a community proclaiming a machine divine.
It may begin with a much less spectacular dispute over whether a sermon was truly preached, whether a ruling was truly issued, whether a commandment was truly performed—or whether the human whose name appears on the act was still its author in any religiously meaningful sense.
The schism has not yet been documented.
Its doctrinal vocabulary is already being written.
References
- Leo XIV. “Meeting with the Clergy of the Diocese of Rome.” Holy See, 19 February 2026. Official record.
- Leo XIV. “Address to the College of Cardinals.” Holy See, 10 May 2025. Official record.
- Egypt’s Dar al-Ifta. “Using AI Applications to Obtain Fatwas.” 2 December 2025. Official ruling.
- Islamic Affairs and Charitable Activities Department, Dubai. “Launching the Project of the Fatwas by 10X.” Event held 29–31 October 2019. Official event record.
- Islamic Affairs and Charitable Activities Department, Dubai. “Artificial Intelligence Supports the Fatwa System in Dubai.” 7 March 2026. Official institutional account.
- Nevins, Daniel. “Halakhic Responses to Artificial Intelligence and Autonomous Machines.” CJLS HM 182.1.2019, approved 19 June 2019. Official PDF.
- Rabbinical Assembly. “Committee on Jewish Law and Standards.” Institutional description.
- Barna Group. “Three Takeaways on How Pastors Can Use AI.” 22 February 2024. Research article.
- Barna Group. “New Research: Pastors Are Using AI More Than You Think.” 15 June 2026; reports data from a December 2025 survey of U.S. Protestant pastors. Research article.