The finding
The sources reviewed show several different relationships: a company-described personal agent, a research and art installation using a Jesus avatar, and an institutional Christian response to AI. These examples justify examining authority as a reporting subject. They do not establish an industry-wide religious project, or show that users generally treat AI as divine.
A tool acts within a task
Start with what a product is asked to do. Summarizing a document is a bounded task: a reader can compare the output with the source. Choosing whom to trust, what to believe or how to live requires a different kind of accountability. The same interface can be used for both. Our reporting therefore asks what the provider advertises, what the system permits and what the user is encouraged to delegate.
The company’s promise and its limits
Meta’s September 8, 2026 Muse announcement describes an agent working on tasks and goals, with user-controlled access and approvals for certain actions. Those are stated design commitments; we have not independently tested them. They are relevant counterevidence against portraying every form of assistance as unlimited autonomy. The next evidentiary step is a documented product test, with the tested version and settings disclosed. Meta announcement
A sacred setting changes the question
Lucerne’s Peterskapelle and HSLU presented Deus in machina, an AI Jesus installation, from August 23 to October 20, 2024. HSLU describes a project investigating spiritual interaction and the limits of trust in a machine. An installation in a confessional is not, by itself, evidence that a church authorized AI to administer a sacrament. The project’s research and art purposes belong in any fair account. Source: hslu.ch
Why representation matters
A text box labeled assistant and an avatar presented as Jesus do not ask readers to interpret the same signs. The latter raises questions about who is speaking, whose teaching the system reproduces and who corrects its errors. This is AI Faith Monitor’s analysis of the setting, not a measured finding about every visitor’s experience. To test an effect, researchers would need an explicit comparison and a clear account of participant selection.
A published Christian response
The Vatican’s January 28, 2025 Antiqua et nova distinguishes AI from human intelligence and argues that technology should serve people rather than replace authentic human relationships. This is a specific Catholic institutional position, not a survey of Christians or evidence of AI-company intent. Readers should consult its wider discussion of benefits as well as risks. Source: press.vatican.va
Four relationships to distinguish
Tool: performs a bounded task. Companion: invites repeated personal conversation. Adviser: recommends a course of action. Authority: is treated as having a decisive right to settle a question. These are working analytical categories, not a scientific scale or an inevitable sequence. A user may reject the role a product invites; a system marketed as a tool may nevertheless be used as an adviser.
How we would investigate influence
For each product, preserve the launch page and actual interface wording. Record defaults, personalization, memory, consent and the ability to challenge an answer. Separate advertised capabilities from independently observed behavior. Ask people what they understood the system to be, rather than deciding from a screenshot that they were deceived. Serious claims about dependence or harm need evidence of the outcome and consideration of alternative explanations.
The missing evidence
This is an analysis of published documents. We have not conducted original interviews, tested Muse, or collected a representative user sample. No claim about the prevalence of spiritual reliance follows from these examples. Before a substantial public investigation, seek company responses, independent human-computer-interaction expertise and theological review; publish material answers and acknowledge nonresponses accurately.
Christian perspective
Our concern is where responsibility and trust are placed. A useful generated explanation can still be wrong. A convincing spiritual voice does not gain pastoral accountability merely by sounding compassionate. Christians should test claims, preserve truthful witness and keep human care and moral responsibility visible. This does not require rejecting every use of AI in education or church administration.
What readers can do now
Ask three concrete questions: Can I check this answer? Who is accountable if it is wrong? Am I asking a tool to supply a kind of judgment it cannot justify? When using AI to study a religious text, return to the text and its context, compare interpretations and involve a responsible human teacher. A company’s name is a reason to ask questions, not an answer to them.
The interface can make a promise the disclaimer retreats from
Imagine two otherwise similar services. The first presents itself as a search tool for a church's public teaching documents. It shows the underlying passage and invites a reader to compare the answer with it. The second uses the same material but introduces itself as a personal spiritual guide whose advice should be trusted. These are hypothetical designs, not descriptions of products we have tested.
Their difference is not reducible to whether the generated sentences are accurate. The second service invites a relationship in which users may attribute a role to the system. A careful evaluation would therefore examine the introduction, name, visual setting, voice, repeated prompts and routes to human support together. A disclaimer could be accurate in isolation while failing to explain the encounter created by the rest of the design.
For a religious institution, this is an organizational decision. The institution chooses whether the interface appears alongside a pastor's message, within educational material or in a space normally associated with confidential care. Each placement can raise different expectations. The institution should ask people to describe those expectations instead of assuming that everyone interprets the design as its creators intended.
A research question is not yet a finding about influence
To examine whether an authoritative presentation changes trust, a study would need to define what it means by trust. Willingness to read an answer is different from willingness to act on it. A favorable comment after a demonstration is different from sustained reliance in ordinary life. Researchers should identify the outcome before collecting reactions.
The comparison matters too. Does a study compare a religiously framed system with an otherwise similar neutral interface, or does it merely ask participants whether they enjoyed one experience? Both may yield useful information, but they answer different questions. A report should describe the actual design, participant selection and limits rather than inflate an interesting response into proof of widespread dependence.
This publication has not performed such a study. Our evidence establishes the existence of particular public presentations and arguments. Our analysis identifies questions that deserve investigation. Keeping those roles separate allows readers to use the material without mistaking a suggested research agenda for completed research.
A counterargument: people can recognize performance
A reader may object that users know a screen is a screen and can engage imaginatively without being deceived. Religious art, drama and educational reenactment can involve representations that participants do not confuse with the represented person. It would be patronizing to assume that every participant in a religiously framed AI installation misunderstands what is happening.
That objection is a reason to study users' interpretations rather than assign them beliefs. It does not remove a publisher's responsibility to explain the system accurately. Different users can bring different expectations, and a design intended as an experiment may be encountered without the introductory context its makers supplied. The fair question is whether the presentation makes its status and limits intelligible.
A Christian evaluation should therefore preserve both human agency and institutional responsibility. Users are not automatically helpless before a symbol, and institutions are not relieved of responsibility because some users understand the symbol critically. Good reporting asks what was presented, what people understood and what followed, treating each as an evidentiary question.
An accountable next step for a ministry
Choose one proposed use and describe its limited function in a sentence a visitor could understand. Identify the source material, the person responsible for checking it, and the route for correcting an answer. Then read the whole presentation from a visitor's perspective. If the visual or verbal promise exceeds the actual provision, revise the promise before expanding the service.
These are recommendations drawn from the analysis, not evidence of the performance of any particular product. A ministry can use them to make its own responsibilities clearer while leaving broader research questions open. The result should be a tool people can evaluate and a human commitment they can locate.
