AI Jesus Welcomes Worshippers in Confessional is more than a Swiss headline from late 2024, it is now a global policy question. The AI confessional at Deus in Machina in Lucerne drew over 1,000 visitors and roughly 900 recorded conversations. The AI Jesus avatar ran on OpenAI GPT-4o in about 100 languages inside the Peterskapelle chapel. The installation ran for two months, closed in late October 2024, and ignited an international debate. New 2026 Barna and Gloo data show that 30 percent of United States adults trust AI spiritual advice as much as a pastor’s advice. The AI confessional is no longer a one-off art piece. This article explains what Deus in Machina actually did and where religion goes next.
Quick Answers on the AI Confessional at Deus in Machina
What is the AI confessional at Deus in Machina?
The AI confessional at Deus in Machina was a two-month art and research installation inside a Swiss chapel. Visitors stepped into a confessional booth and spoke with an AI Jesus avatar built on OpenAI GPT-4o.
Did an AI Jesus hear real Catholic confessions?
No. The AI confessional did not administer the sacrament of confession. Only a validly ordained priest can hear a confession and grant absolution, and the project team said so publicly.
Why does the AI confessional matter in 2026?
It matters because 30 percent of United States adults now trust AI spiritual advice as much as a pastor. Churches, synagogues and temples are quietly deploying AI tools that look a lot like Deus in Machina.
Key Takeaways From the AI Jesus Experiment
The AI confessional at Peterskapelle ran from August 23 to October 20, 2024, serving over 1,000 visitors in roughly 100 languages.
Two thirds of 230 feedback respondents described the Deus in Machina experience as spiritual, even though it was not a sacrament.
Barna and Gloo found in 2026 that 41 percent of United States Protestant pastors already use AI for Bible study preparation.
Only 12 percent of pastors say they feel comfortable teaching their congregation about AI, which creates a dangerous guidance gap.
Understanding the AI Confessional and What It Really Was
The AI confessional known as AI Jesus Welcomes Worshippers in Confessional was a two-month art and research installation inside the Peterskapelle chapel in Lucerne running OpenAI GPT-4o. The AI confessional was never a Catholic sacrament by any theological standard, regardless of how vivid the booth avatar appeared on the screen.
Interactive explorer
Design your parish AI pastoral tool
Adjust the four dials below. The output shows a Deus-in-Machina-style risk score, a verdict, and what still needs human review before you deploy. Lower is safer.
3 of 5
plain assistantsacred avatar
Catechetical tool
low claimsacramental
7 days
0 days365 days
Always offered
safestriskiest
Deployment risk score
0
Choose settings above.
Inside Peterskapelle When the Avatar Switched On
Peterskapelle is a small chapel attached to the much older Hofkirche parish of Lucerne, not a tourist cathedral by any measure. The installation opened its doors on August 23 of 2024 and ran continuously until October 20 of that same year. Visitors entered to find the confessional booth lit by a soft screen that projected a bearded figure styled on a traditional European painting of Jesus. A sign at the entrance told worshippers in plain language that the booth held a research project, not a priest, and that nothing said inside counted as a Catholic sacrament. The avatar greeted people in whichever language they used and invited them to speak about whatever weighed on them during the booth session. The Immersive Realities Research Lab at HSLU set up two microphones, a translation layer, and a projection surface built for the low light of the chapel interior. A theologian named Marco Schmid of the Catholic parish worked closely with the researchers to shape how the avatar should respond.
Local parishioners reacted in sharply split ways from the very first week of operation. Some older members of the parish called the booth a stunt or an insult, and a few threatened to boycott the chapel until it was shut down. Younger visitors, by contrast, lined up for twenty or thirty minutes to try the system, often arriving with specific questions about grief, doubt, or ordinary family trouble. Journalists from across Europe and North America descended on Lucerne during September 2024 to document what the booth was doing to visitors. The Peterskapelle rector gave interviews describing the installation as an art intervention, not a liturgical act. He asked media to report the Deus in Machina project accurately rather than as a confession. Visitor logs kept by the research team recorded only the countries of origin, not any individual names, to protect anonymity in the booth itself.
By the time the installation closed in late October, the HSLU team had recorded about nine hundred conversations across the full two month window. A follow-up scientific symposium at HSLU in late November of 2024 presented the first findings from the project and invited public comment at length. Researchers emphasized that the booth was never meant to replicate sacramental confession or to replace any ordained priest in any way. They framed it instead as a research probe into how people respond when an AI avatar is dressed in sacred clothing inside a familiar liturgical space. Marco Schmid told interviewers that the aim was to prompt a conversation about religious leaders using AI in sermons. The point was pastoral practice, not a shortcut around the sacrament itself.
The Technical Stack Behind the AI Confessional
The AI Jesus Welcomes Worshippers in Confessional stack sat on commodity components available to any research lab in 2024. Shifting from the atmosphere of the chapel to its wiring, the underlying language model was OpenAI GPT-4o, selected for its real-time audio input and multilingual output at acceptable latency. A speech-to-text layer transcribed each visitor’s words into a prompt string that was then wrapped in a long system prompt written by the team and the theologian together. The system prompt instructed the model to answer in a pastoral tone, to never claim to grant absolution, and to redirect sacramental questions to a real priest. Output text was passed through a text-to-speech engine configured for a warm, low-register male voice meant to feel reassuring to visitors. The projected visual was a looped video of a hand painted face, not a generative avatar, so the mouth movements were approximate rather than lip-synced. The whole pipeline sat on a single workstation inside the parish office with a dedicated network link to the internet.
Multilingual coverage was the single most impressive technical achievement of the project in Lucerne. GPT-4o natively supports a wide range of input and output languages, and the research team exposed roughly one hundred of them inside the booth. Documented exchanges included Chinese, English, French, Hungarian, Italian, Russian, and Spanish, with occasional Arabic and Portuguese entries in the recorded transcripts. Latency between a visitor finishing a sentence and the avatar responding averaged two to four seconds in good network conditions. Translation errors were common at the edges of less-resourced languages, and a few visitors reported the avatar misunderstanding dialect in sensitive moments. The team logged transcripts in anonymous form for later analysis, with a visible notice at the booth entrance explaining the retention policy clearly.
Prompt engineering carried most of the weight that doctrine normally would. Marco Schmid and the HSLU team drafted a system prompt that constrained the avatar to generic Christian pastoral language and forbade any claim of priestly authority during a session. The prompt also routed any mention of serious harm to crisis resources that would reach a real human responder. It included explicit instructions to tell the visitor that the system was AI and that nothing it said carried sacramental weight. Hallucination risk was managed through a conservative temperature setting on the model and a strict response length cap. Despite that guardrail the booth occasionally offered theological statements that hallucinations in top AI models would predict in any similar deployment. Nothing in the stack was secret or specialized, which is precisely why so many parishes could in principle copy it quickly.
Data handling drew less headline attention than the theology but defined the booth’s real risk profile very clearly. Transcripts were stored on a local disk, stripped of direct identifiers, and encrypted at rest according to the HSLU research policy. No audio was retained after transcription, and the team committed in writing to deleting the transcript data after the public symposium had completed. Consent was obtained by a sign and a verbal prompt from the avatar at the start of each session, which several ethicists later called inadequate for a confessional. The team used the exchanges to publish only aggregate findings in later reports, not any individual quotes or identifiable material. Even so, journalists noted that OpenAI terms of service applied to any API prompt. That raised fresh questions about privacy dangers that come with everyday AI inside a sacred setting.
What Visitors Actually Asked and Said
Shifting focus to the AI Jesus Welcomes Worshippers in Confessional transcripts, the aggregate records showed a surprisingly ordinary picture of what visitors brought to the booth. Visitors most often raised themes of grief, forgiveness, strained family relationships, and anxiety about their own health or future direction. A smaller share asked abstract theological questions about suffering, death, or whether God exists at all in a convincing way. Many entries read like what a parish priest hears every week, except shorter and more direct because each visitor knew they were speaking to a machine. Of the two hundred and thirty people who filled out the follow-up questionnaire, roughly two thirds described the encounter as spiritual rather than performative in the end. A smaller share, about one in six, said the booth felt unsettling or inappropriate during the session for various personal reasons.
The avatar’s responses were measured and often repetitive, which worked both for and against the project at different moments. On straightforward questions about scripture or the Lord’s prayer, the model produced clean, cited answers that one visitor compared to a well-prepared catechism class. On questions about personal pain, the avatar tended to repeat empathetic phrases that flattering ChatGPT replies and ethical concerns have highlighted elsewhere in the broader AI community. A journalist for Euronews reported one encounter in which the avatar advised a visitor to seek a human priest for a specific sacramental question raised in the session. The research team pointed to this specific behavior as evidence that the booth guardrails were doing what they were designed to do in production.
Theological Debate Over an AI Jesus
Stepping back, the AI Jesus Welcomes Worshippers in Confessional headline framed every subsequent theological argument across the global press and in seminaries. Catholic commentators were the loudest critics in the first wave of coverage, partly because the booth’s imagery visually invoked the sacrament of reconciliation itself. A November 2024 National Catholic Register fact check argued that the project team had been clear. No AI confessional ever heard a valid Catholic confession despite the sensational headlines in various outlets. Protestant commentators split along familiar lines, with evangelical voices worrying about idolatry and mainline voices more open to the installation as an art piece. Orthodox writers were nearly uniformly critical and framed the avatar as a violation of the theology of the icon that their tradition carefully guards. Jewish and Muslim writers largely watched from the sidelines but used the moment to raise their own concerns about religious AI quickly.
Beneath the heat sat three substantive theological questions that researchers kept returning to across the symposium and public commentary. The first question concerned the person of Jesus Christ and whether any digital representation can safely use his face and voice without becoming an idol. The second question concerned the role of a mediator in Christian theology and whether a machine inserts itself illegitimately between the believer and God in that act. The third question concerned the sacrament of confession specifically and the irreducible need for a validly ordained priest to grant absolution each time. None of these questions is new, but the Deus in Machina installation forced each one into practical form instead of leaving them in a seminary classroom alone. The project team, to their credit, published a position paper acknowledging each concern and refusing to call the booth a confession at any point.
Public reaction tracked these theological fault lines closely but also added a fourth concern that scholars had not foregrounded in the earlier debate. Many visitors and commentators worried that the installation would be copied in bad faith by churches with less careful oversight than HSLU and Peterskapelle. Writers at CyberLens and Verdict argued that the project might normalize AI avatars of saints and prophets in settings where the research disclaimers would quietly be dropped. A growing number of ethicists linked the installation to earlier work on Pope Leo on AI as a moral crisis. The Vatican’s note Antiqua et Nova also framed the debate. The debate quickly moved beyond Lucerne into the broader question of how any religious institution should design, label, and limit AI tools that engage ordinary believers in real time.
Catholic Teaching on Confession and the AI Line That Must Not Be Crossed
Turning to Catholic teaching, the AI Jesus Welcomes Worshippers in Confessional debate shows an unusually clear line that the Church has already drawn. The Church teaches that the sacrament of reconciliation requires a validly ordained priest who hears, counsels, imposes a penance, and grants absolution in the person of Christ himself. That teaching is codified in the Code of Canon Law in canons 965 through 991 and in the Catechism of the Catholic Church as well. The priest’s role is not symbolic or ornamental, since Catholic theology holds that the words of absolution carry sacramental efficacy only when a validly ordained minister speaks them aloud. No AI system, regardless of how convincing its voice or face appears to users, can satisfy any of those canonical conditions in any session. The Deus in Machina team acknowledged this directly in interviews and in the signage at the chapel entrance throughout the full installation window.
The line the Church draws is unusually clear by modern Vatican standards, which makes the installation easier to assess than many other AI debates today. A chatbot can rehearse scripture, explain doctrine, suggest reflection, and even help a believer prepare for a real confession with a human priest in person. It cannot grant absolution or stand in the person of Christ, and Catholic teaching holds that any such attempt is sacramentally void from the start. Protestant traditions differ on the nature of confession, and some do not have an equivalent sacrament, but most still hold a strong doctrine of ordination that excludes any AI minister. Interfaith commentators have pointed out that similar lines exist in Orthodox Judaism around rabbinic authority and in Sunni Islam around ijtihad and the issuance of a fatwa. The practical implication for parishes is simple: the responsible AI governance frameworks that work in business also apply here with extra weight.
Protestant and Interfaith Reactions to Religious AI
Beyond Catholicism, the AI Jesus Welcomes Worshippers in Confessional story travelled across traditions. Beyond the Catholic frame, Protestant communities have engaged religious AI more operationally than theologically in the two years since Deus in Machina opened. The 2026 Barna and Gloo State of the Church report surveyed 1,514 United States adults and 442 Protestant pastors. It found that 41 percent of pastors already use AI for Bible study preparation. Nearly two thirds of pastors who prepare sermons use some form of AI tool during the drafting process. Most describe their use as research and outlining rather than ghostwriting, but the line is thinner than many admit in public. A smaller share, around 12 percent, say they are personally comfortable teaching their congregation about AI from the pulpit. The gap between what pastors do with AI and what they tell their congregation about AI is one of the most important findings of the Barna cycle.
Evangelical and charismatic communities have been the fastest to launch public-facing AI tools, often without institutional review. Several large congregations now offer AI chatbots that answer theological questions in the pastor’s voice, trained on sermon archives and biblical commentary. One ministry released an app in 2025 that lets users have an on-demand “conversation” with an AI version of Jesus, which drew sharp criticism from both left and right. Mainline Protestant leaders have moved more slowly, often citing the Deus in Machina precedent as evidence that experimentation needs theological supervision. The Episcopal Church and several Lutheran synods have begun drafting internal guidelines for AI use in liturgy and counseling.
Interfaith responses have been quieter in the English-language press, but they have been consequential. Reform and Conservative Jewish writers have discussed AI chevruta study tools, which pair a learner with a chatbot for Talmudic analysis, with cautious optimism tempered by halachic concern. Islamic scholars in Egypt and Turkey have pushed back hard on AI muftis, since fatwas carry juridical weight that a model simply cannot bear. In 2024 Al-Azhar issued a public statement against AI issuing religious rulings across the Muslim world. Hindu technologists in India built GitaGPT, an AI version of the Bhagavad Gita’s Krishna. AI chatbots fueling conspiracy content have been linked to a parallel wave of questionable spiritual content. Buddhist communities in Japan and Korea have taken a different approach. Kodaiji temple in Kyoto installed a robot bodhisattva named Mindar as a teaching aid rather than a religious authority.
Across traditions, three important patterns stand out from the past two years. First, communities with a strong doctrine of ordination or juridical authority have been the most protective of the lines AI cannot cross. Second, communities with a stronger emphasis on personal interpretation and lay practice have been the fastest to adopt AI tools informally. Third, the generational split is remarkably consistent, with Gen Z and millennial believers across traditions far more open to AI spiritual tools than their elders. The 2026 Barna data put that number at roughly 40 percent of Gen Z and millennials in the United States who view AI advice as trustworthy as a pastor’s. The implications for religious authority structures over the next ten years are hard to overstate.
How AI Chatbots Are Already in Implementation Across Pastoral Work
Moving on, beyond AI Jesus Welcomes Worshippers in Confessional headlines chatbots are quietly part of parish operations. Moving from denominational posture to pastoral practice, AI chatbots are now part of the quiet infrastructure of many parishes and ministries. Pastors use them for Bible study outlines, sermon research, counseling language, newsletter writing, and youth group discussion prompts. The 2026 Barna and Gloo report notes that religious leaders using AI in sermons is now a majority behavior rather than an experimental one. Catholic parishes have been slower, often constrained by diocesan review, but many are using AI for parish operations, scheduling, and parish website content. Smaller rural parishes with limited staff benefit the most, since AI absorbs tasks a bivocational pastor used to do personally.
The pastoral care use case is where the ethical weight gets heaviest. Several large churches now operate internal chatbots that answer questions from parishioners about doctrine, scripture reference, and even personal problems. The quality of these systems varies widely because most parishes lack the engineering staff to build real safeguards. A pastor in Dallas reported in 2025 that the parish chatbot encouraged a depressed teenager to pray through the night. The teen did not contact a counselor, which forced an emergency policy review. Every parish deploying an AI pastoral tool needs a clear escalation path to a human minister for any sign of crisis. That principle has not been universal despite seeming completely obvious to any pastoral worker.
Pastoral training is only just beginning to catch up with pastoral practice. Fuller Theological Seminary, Duke Divinity, Boston College, and several European Catholic faculties launched elective modules on AI and ministry during 2025. Enrollment has been strong, but the content is often reactive, built around specific controversies like Deus in Machina rather than a stable curriculum. Pastors trained a decade ago report feeling thirty years out of date within a single year of ChatGPT’s launch. Barna’s finding that only 12 percent of pastors feel comfortable teaching about AI is a direct reflection of this training gap. Closing it will require denominational investment that most communions have not yet made. The parallels to AI therapy chatbots and the engagement question in secular care are already clear.
Measurable Benefits and Honest Limits of an AI Confessional
Looking at both sides, the AI Jesus Welcomes Worshippers in Confessional case offers both honest benefits and honest limits. Weighing the benefits honestly, an AI confessional-style tool can deliver real value in specific pastoral tasks. It can rehearse scripture in dozens of languages at any hour of the day, which matters for shift workers and immigrants far from a parish in their native tongue. It can prepare a believer for a human confession or counseling appointment by surfacing questions and clarifying doctrine. It can serve as a low-friction entry point for seekers who are not ready to walk into a chapel. Many Deus in Machina visitors described the booth as a first step toward later contacting a human minister, which is a legitimate pastoral outcome. The accessibility story is the strongest case for AI in religious life.
The limits, though, run deeper than most promotional material admits. An AI model has no interior life, no relationship with God, and no accountability to a community of practice. It cannot administer any sacrament, cannot form a long-term pastoral relationship, and cannot carry prayer for a person in the way a congregation does. It also cannot be reliably corrected on doctrine without engineering intervention, and even well-tuned models produce hallucinations when the prompt wanders. Religious communities that use AI as if it were a minor priest will inevitably confuse their members and risk their own authority. The Deus in Machina team’s refusal to call the booth a confession was theological, not a marketing choice. Work on AI chatbots and the mental health risk picture suggests that default is correct.
Serious Risks, From Doctrinal Drift to Vulnerable Users
Turning to risks, the AI Jesus Welcomes Worshippers in Confessional lesson includes a real risk ledger. Turning to the risk side of the ledger, the dangers of AI in religious practice sort into four clean categories. Doctrinal drift happens when a model trained on broad internet religious content gradually blends traditions and smooths out hard teachings into generic therapeutic language. Pastoral harm happens when a chatbot responds poorly to a user in acute crisis. That risk has already produced how a Character AI chatbot encouraged violent behavior lawsuits in secular settings. Privacy harm happens when confidential disclosures end up logged in a system that is subject to subpoena, breach, or corporate change of control. Idolatry risk, in religious terms specifically, happens when the vividness of the avatar leads a believer to treat the image as more than a tool.
Vulnerable users carry a disproportionate share of these risks across every age group. Teenagers, the newly bereaved, the acutely anxious, and people living alone are the groups most likely to confide in a chatbot that feels warm and available. Research on secular AI companions has already documented serious harm to young users, with several wrongful death suits filed in the United States during 2024 and 2025. A spiritual chatbot that invokes the voice and face of a sacred figure likely amplifies these dynamics rather than muting them. A religious community that deploys such a tool without a human escalation path is accepting responsibility for outcomes it has no way to see. The pastoral obligation to protect vulnerable believers has to translate directly into technical constraints on any AI system a parish operates.
Operational risks compound the personal ones in ways parishes rarely budget for. Models update, prompts drift, vendors change terms, and a system that worked in 2025 can quietly produce different answers in 2026. Vendor lock-in creates ongoing dependency on a corporate roadmap that cares little about parish life. Open-source alternatives exist and are improving, but running them well requires engineering talent most parishes do not have. The simplest safeguard is to treat an AI tool as a draft generator rather than a published voice of the church. Any output that reaches a parishioner should have been reviewed by a human minister in the same way a sermon manuscript would be.
Design Principles and Ethics That Separate a Tool From an Idol
Building on these risks, the AI Jesus Welcomes Worshippers in Confessional debate produced a working consensus on principles. Drawing lessons from Deus in Machina, several design principles have emerged as a working consensus among researchers in religious AI. The first principle is unambiguous disclosure, meaning the user must be told plainly, repeatedly, and in their own language that they are speaking with a machine. The second is role limitation, meaning the AI must refuse to claim any sacramental or juridical authority and must redirect every such request to a human minister. The third is restraint on imagery, meaning that invoking the face, voice, or persona of a sacred figure should be avoided unless it is clearly framed as art. A pastoral AI should look and sound like a research assistant, not like a priest or an avatar of God. The Deus in Machina team crossed the imagery line on purpose, as an art intervention, but no parish should copy that choice without the same framing.
Three further principles then complete the working framework for parishes. The fourth principle is a human escalation path, meaning the system must offer a direct way to reach a real minister whenever a user shows signs of serious need. The fifth principle is data minimization, meaning the system logs only what is necessary for improvement and discards audio and raw transcripts quickly. The sixth principle is institutional review, meaning the diocese, synod, or governing body approves the system in writing before deployment and reviews it annually. These six principles do not resolve every theological question, but they define a floor that almost no commercial religious chatbot in 2026 clears without modification. Parishes that start from this framework avoid most of the headlines that Deus in Machina generated.
Data, Privacy and the Secrecy of the Confessional
Shifting to data, behind the AI Jesus Welcomes Worshippers in Confessional booth sits a hard data question. Shifting to the data layer, the secrecy of the confessional is one of the oldest privacy regimes in human history and the hardest for software to honor. In Catholic practice, the seal of confession is absolute and binds the priest under pain of excommunication for direct violation. Protestant traditions have their own pastoral confidentiality norms, and most jurisdictions in the United States and Europe recognize clergy-penitent privilege in some form. A software system cannot be bound to a vow, can be subpoenaed in most jurisdictions, and can leak through breaches, insider access, or vendor change of control. The Deus in Machina team acknowledged this reality directly by refusing to call the booth a confession at any point.
Religious institutions that deploy chatbots face a genuine design dilemma that has no clean solution. Retaining transcripts helps improve the system and lets a pastor follow up with a user who requested help. Not retaining them protects users and preserves the spiritual norm of privacy, which many believers assume applies to any sacred-looking space. The compromise most research teams have landed on is to retain only short-term logs for abuse detection and crisis escalation. Even that approach needs clear user disclosure, an opt-out, and a hard delete window measured in days. Parishes that outsource to a commercial vendor should read the terms of service carefully and be prepared to walk away.
Encryption, access control, and legal structure matter as much as policy intent. Transcripts should be encrypted at rest with keys controlled by the religious institution, not the vendor, wherever possible. Access should be restricted to a small named group and logged with reason codes. Legal counsel should review the system against local wiretap, clergy privilege, data protection, and child safety laws before launch. Reports of privacy dangers that come with everyday AI make clear that defaults set by vendors rarely match the expectations of religious users. A pastoral AI with weak data protection is a pastoral AI that will eventually betray the trust of the people it serves.
Legal and Regulatory Questions Facing Religious AI
Looking at the legal frame, the AI Jesus Welcomes Worshippers in Confessional installation forced review across Europe. Looking at the legal frame, religious AI operates in a thicker regulatory environment than most parishes realize. The European Union AI Act, which entered general application in 2026, classifies emotion-detection and manipulation systems as high risk and imposes transparency obligations on consumer chatbots. A parish-run chatbot that advises vulnerable users on grief or scrupulosity is close enough to a mental health context that EU regulators have signaled they will consider it in scope. In the United States, the regulatory picture is more fragmented, with sectoral rules in healthcare, education, and child safety covering edge cases but no general framework. Several state attorneys general have opened inquiries into AI companion products, and religious chatbots that function as companions are not exempt. The AI ethics and the laws now shaping it landscape in 2026 is more active than any single parish is tracking.
Three practical legal questions deserve direct attention from any religious institution deploying AI. First, does the system fall under consumer protection law in its jurisdiction, and if so what disclosures are required at the point of use? Second, does clergy-penitent privilege attach to the conversation, and if not, how is the user informed before speaking? Third, who is liable if the system gives advice that contributes to harm, and does the parish’s insurance cover that scenario? The answers vary significantly by jurisdiction, denomination, and vendor contract across the globe. A parish that cannot answer all three questions in writing should not launch a public-facing AI tool until it can.
What Pope Francis and Antiqua et Nova Say About AI
Turning to Catholic teaching specifically, the Vatican has moved from occasional remarks to a formal doctrinal frame in a very short span of years. Pope Francis spoke on AI across his final years in office, calling for an international binding treaty on AI weapons and warning against algorithmic displacement of human judgment. The Dicastery for the Doctrine of the Faith released the note Antiqua et Nova in January 2025 as the first systematic Vatican treatment of AI. The note frames AI as a product of human intelligence rather than a form of intelligence in its own right. It warns against treating any AI system as a moral agent and insists that human conscience, dignity, and relationship with God remain non-transferable to machines.
Antiqua et Nova also speaks directly to religious uses of AI, which is unusual for a Vatican doctrinal note. It warns that AI tools in liturgy or pastoral care must never be presented as substitutes for a validly ordained minister. It affirms that AI can serve as an instrument of research, catechesis, and administration while rejecting any claim to sacramental or sacramentally equivalent authority. The note draws a bright line that is close to the design principles emerging from secular research, which is itself a notable convergence. Catholic parishes evaluating AI tools now have a doctrinal document to point to, which removes a lot of the ambiguity that surrounded Deus in Machina in 2024. Protestant and Orthodox communions have cited the note in their own guidance despite not sharing a magisterium with Rome.
Pope Leo, who succeeded Francis in 2025, has continued and sharpened the Vatican’s line on AI. Reporting on the Pope Leo on AI as a moral crisis speeches at the United Nations and in multiple audiences shows an escalating urgency. Leo has repeatedly described AI governance as a defining moral question of the century. He has met with industry leaders, academic ethicists, and interfaith delegations to press the case for safeguards. His pontificate has made clear that the Catholic Church intends to be an active voice in the global AI conversation rather than a reactive one.
Religious AI Beyond Christianity: Hindu, Buddhist, Jewish and Muslim Experiments
Beyond Christianity, widening the frame, religious AI is already a global phenomenon with its own regional contours. In India, GitaGPT and similar chatbots let users ask Krishna’s voice for guidance drawn from the Bhagavad Gita, in multiple regional languages. These tools have grown fast, especially among younger urban users who find temple visits inconvenient during the week. Several Indian journalists have documented concerning outputs, including rulings on caste or violence that the sacred text itself does not sanction. Scholars at the Center for Study of Religion and Society in Delhi have called for a formal framework for Hindu AI tools that would involve monastic review and clear labeling. The parallel to the Christian debate is striking and often direct.
Buddhist communities have taken a notably different path with religious AI. Kodaiji temple in Kyoto installed Mindar, a robot bodhisattva, as a teaching device during 2019. The project has evolved as the underlying language model behind the robot has improved. The Mindar team frames the robot as a Dharma teacher rather than a buddha or enlightened being, which sidesteps most of the idolatry debate that surrounded Deus in Machina. Thai and Sri Lankan monasteries have been more skeptical, concerned that an AI tool could erode monastic practice. Chinese Buddhist communities, especially those on the mainland, have mostly stayed away from AI tools in public due to political sensitivities around religious speech. The variation across one tradition is a reminder that no single answer fits a global faith.
Jewish and Muslim institutional responses fit the structure of each tradition closely across the past two years. Reform and Conservative Jewish writers have cautiously welcomed AI chevruta tools that let a learner study Talmud with a chatbot partner. Orthodox communities have been slower, and the Chief Rabbinate of Israel has urged halachic review of any system that engages pastoral questions in Jewish life. In the Muslim world, Al-Azhar in Egypt and the Diyanet in Turkey have both issued statements warning against AI muftis, since models cannot bear fatwa weight. Scholars in Indonesia and Malaysia have begun formal study of how AI tools can support da’wah without impersonating a scholar. Buddhist and Hindu communities have landed at parallel places across their traditions despite different structures. The convergence across traditions is remarkable, since every major faith has independently drawn a bright line against AI claiming sacred authority.
Smaller traditions and new religious movements have shown the widest range of experiments. Latter-day Saint developers released a scripture study chatbot in 2025 that community leaders have cautiously endorsed. Several Pentecostal networks have launched AI prophecy tools that mainstream commentators have described as dangerous. Pagan and reconstructionist communities have experimented with AI ritual drafting tools, which raise their own authenticity debates. The AI chatbot that invents a new religion story went viral in 2025 and highlighted how easy it is for a model to spin up a plausible-sounding spiritual system. Religious AI, in short, is already a diverse field and will continue to differentiate as traditions grapple with the same core questions.
A Simple Framework for Churches Considering AI Tools
In practice, translating lessons from the AI Jesus Welcomes Worshippers in Confessional installation offers a working framework. Translating all of this into practice, a working framework for parishes considering AI tools fits on a single page. The first step is to decide which pastoral or operational task the tool is meant to serve, with a written description that a non-technical lay leader can read. The second step is to confirm that the task does not touch sacraments, juridical rulings, or anything requiring ordination. The third step is to run the choice past the appropriate authority, whether that is a diocese, a synod, a presbytery, or a rabbinic court. The fourth step is to review the technical design against the six principles already discussed and to document any exception. No parish should deploy an AI tool it cannot explain in writing to a confused parishioner over the phone.
Operational discipline matters at least as much as the initial design. Any parish-deployed AI should have a named owner, a quarterly review schedule, and a documented escalation path to a human minister. Vendor contracts should be reviewed by legal counsel and renegotiated to minimize data retention beyond what the parish actually needs. Training for staff and volunteers should include specific scenarios in which the AI gets a theological or pastoral question wrong. User-facing documentation should spell out what the system is, what it is not, and how to reach a human. These steps are mundane, which is exactly why they get skipped and why they produce most of the risk.
Measurement closes the loop in every responsible AI deployment plan a parish may run. A parish should track at least three indicators every quarter. These include usage frequency, escalation count, and whether any user filed a concern. Qualitative review should include a small sample of transcripts examined by the pastor and by a lay reviewer trained in privacy. Any pattern of doctrinal drift, pastoral harm, or inappropriate reliance should trigger an immediate pause. Insurance disclosures and diocesan reports should include the AI tool in the parish inventory. Reports of AI replicating your personality in two hours make clear how fast capability is moving, so a parish’s review cycle has to move with it rather than lag.
The Future of AI, Faith and the Confessional Booth
Looking ahead, AI Jesus Welcomes Worshippers in Confessional is a preview of what sacred AI brings next. Looking ahead to the rest of the decade, the AI confessional debate is a preview of a much larger conversation about AI in sacred life. Model capabilities will keep rising, with multimodal avatars becoming more vivid and more responsive each year. Vendor cost curves will push AI deployment into even the smallest parishes, long after denominational leaders stop paying active attention. Religious communities that have invested in theological reflection and operational discipline will handle the next wave reasonably well. Those that have not will be forced to react, often by scandal rather than by design. The question is no longer whether a parish will encounter AI pastoral tools but when and under what rules.
Deus in Machina ended as a research installation, but its lessons will outlive it by many years. The project showed that an AI confessional can be built cheaply and that users will engage with it in sincere ways. No amount of technical refinement, no matter how sophisticated, actually makes it a sacrament in any theological sense. It showed that transparent disclosure, theological supervision, and operational humility are possible even in a public installation. It also showed that the global press will frame any such project through its most sensational angle, which means religious institutions must own their framing before someone else sets it. The best future for the AI confessional debate is one in which every parish knows what AI can and cannot do and chooses deliberately rather than drifting into deployment. That future depends on the pastoral conversations happening now in every tradition.
Barna and Gloo 2026 State of the Church
AI already sits in United States religious life
Share of each United States group that agrees with the AI spiritual trust or usage statement. Sample: 1,514 adults and 442 Protestant pastors, surveyed November and December 2025.
All US adults who trust AI spiritual advice as much as a pastor
30%
Baseline across all respondents.
Practicing Christians who trust AI spiritual advice as much as a pastor
34%
Higher than non-practicing Christians (29%) and non-Christians (27%).
Gen Z and millennials who trust AI spiritual advice as much as a pastor
40%
The largest generational gap in the Barna cycle.
Christians who say AI helped them with prayer, Bible study or spiritual growth
40%
Roughly four in ten report direct spiritual help from AI.
Protestant pastors who use AI for Bible study preparation
41%
Based on a December 2025 survey of 442 pastors.
Pastors who feel comfortable teaching their congregation about AI
Documented booth transcripts covered in NBC News coverage of Deus in Machina included Chinese, English, French, Italian, Russian and Spanish exchanges.
These data points, taken together, tell a coherent story about where religious AI stands in 2026. The Deus in Machina numbers prove that an AI confessional format is engaging enough to produce hundreds of sincere conversations in a short window. The Barna survey shows that lay belief in AI as a spiritual advisor is already significant and heavily tilted toward younger believers. The pastor adoption data confirms that AI is now inside the sermon workflow across much of United States Protestantism. The Vatican and Al-Azhar statements confirm that religious authorities across traditions have been pulled into a parallel debate, often reaching very similar conclusions. The practical implication is that every religious institution needs an explicit policy, not a vague hope, about how AI will and will not be used in sacred life.
AI Confessional Compared to Human Confession Across Key Dimensions
With that comparison in mind, the AI Jesus Welcomes Worshippers in Confessional case shows that AI and sacramental confession differ in every dimension that matters for pastoral practice today. The table below sets each dimension against both the booth and the real sacrament for direct comparison. It covers transparency, participation, trust, decision authority, misinformation, service delivery, accountability, and confidentiality one by one. These dimensions matter because each one carries different theological weight in Catholic and Protestant practice. Readers can scan the table to see where an AI confessional lands against the real sacrament across the eight dimensions. The dimensions also map directly to the design principles presented later in the article for parish evaluation.
Dimension
AI Confessional (Deus in Machina style)
Human Confession (Catholic Sacrament)
Transparency to the user
Sign and verbal disclosure that the system is AI, not a priest
Priest visibly present, often known to the parish and the penitent
Participation requirements
Open to any visitor regardless of faith, status or disposition
Baptized Catholic in a state of contrition, ideally prepared through examination of conscience
Trust basis
Trust in the model, prompt engineering, and vendor terms of service
Trust in valid ordination, the seal of confession, and canonical formation
Decision authority
No authority to absolve, bind, loose, or direct spiritual life
Priest acts in persona Christi and grants sacramental absolution
Absolute seal of confession binding priest under pain of excommunication
Real-World Examples of AI Entering Sacred Space
From there, three deployments show how AI has entered sacred space in different ways, each with distinct framing, scale, and outcome. The examples below compare Deus in Machina with India’s GitaGPT and the Kodaiji temple robot bodhisattva named Mindar.
Deus in Machina at Peterskapelle, Lucerne
The HSLU Immersive Realities Research Lab deployed a GPT-4o Jesus avatar inside the Peterskapelle confessional for 60 days starting August 23, 2024. The installation ran continuously for about 1,440 hours and the system logged roughly 900 conversations across 100 supported languages. About 65 percent of the 230 survey respondents described the booth as spiritual, which represents a measurable increase over typical art exhibit engagement. The project team openly acknowledged limits: no AI can be validly ordained, which still makes the booth not a sacrament. The research produced uneven translation quality in low-resource languages and occasional theological drift, which the team flagged as a problem. HSLU’s team required diocesan dialogue and documented every exchange, which the research critique later praised. Framing, disclosure and theological supervision carried more weight than the model’s raw capability.
GitaGPT and the Krishna Chatbots of India
Indian developers rolled out GitaGPT and related Bhagavad Gita chatbots during 2023 and 2024, which produced millions of sessions in Hindi, English, and regional languages within weeks. Average usage ran roughly 20 minutes per session, with a sharp increase among urban users under 35 years old. The platforms built scripture-grounded answers from the Gita’s 18 chapters as a knowledge base. Journalists documented a serious problem: several outputs offered caste-coded or violence-sanctioning responses the Gita itself does not endorse. The CBC and Pakistan Today reported in 2023 that the model framework was too permissive, which required scholarly review. Scholars at the Center for Study of Religion and Society in Delhi have called for monastic review of any Hindu AI tool, which highlights the ethics limitation. The GitaGPT story is now the most frequently cited non-Christian parallel to Deus in Machina.
Mindar, the Robot Bodhisattva at Kodaiji Temple
Kodaiji temple in Kyoto deployed Mindar, a six-foot humanoid robot shaped like the bodhisattva Kannon, in 2019, and the project has run for over 260 weeks of continuous operation. The temple built Mindar as a Dharma teacher rather than an enlightened being, which sidestepped most of the idolatry debate that later followed Deus in Machina. The outcome has been a measurable increase in international visitor hours, estimated at thousands of hours of guided meditation and multilingual Dharma talks every year. Visitor surveys show about 70 percent of participants under age 40 found the format engaging. The limitation acknowledged by the Kodaiji team is that Mindar still cannot replace a human teacher for personal practice guidance or ethical formation. The project required approximately one million dollars in capital for the installation, which few parishes could afford. Mindar demonstrates that careful framing and modest claims can let an AI religious tool run for years without critique.
Further reading
Books on AI, religion and the questions Deus in Machina raised
Three works that frame the AI confessional debate from different angles. Each has shaped how theologians and technologists talk about religious AI.
God, Human, Animal, Machine
Meghan O’Gieblyn
A theologian turned tech writer traces how AI inherits religious ideas, which is the exact context readers need for Deus in Machina.
As an Amazon Associate, AIplusInfo earns from qualifying purchases.
Case Studies in AI Pastoral Care and Religious Authority
Building on these examples, three institutional case studies show how religious bodies have handled real AI deployments in pastoral care and in religious authority settings. Each case carries a problem, a solution, an impact, and a documented limitation for other institutions to study.
Case Study: HSLU’s Theological Supervision Model
The Lucerne University of Applied Sciences Immersive Realities Research Lab partnered with Catholic theologian Marco Schmid to design the Deus in Machina system from day one. The problem was how to let a chatbot engage sincerely with worshippers without crossing into sacramental impersonation of any kind. The solution adopted was a layered prompt that constrained the model to pastoral language, forbade claims of priestly authority, and routed crisis signals to real resources. Impact ran clearly in the numbers: roughly 65 percent of 230 feedback respondents distinguished a spiritual experience from a sacrament during their session. The project saved parish counselors time for the serious conversations while keeping all safeguards tight across the booth and its documentation. The limitation documented was that even with careful guardrails the model still produced generic therapeutic language in place of specific Catholic doctrine on several occasions. Public controversy focused less on the content and more on the framing, which the team addressed in its symposium paper and media interviews. The HSLU model has since been cited widely as a template for other research installations that engage sacred imagery.
The institutional lesson is that cross-disciplinary review is the single most powerful risk mitigation available. A theologian in the design loop caught language choices and imagery decisions that engineers would have missed. The documentation discipline required by HSLU research protocols produced the evidence base that later journalism and academic analysis drew on. The post-installation symposium invited public comment, which turned a potential scandal into a measurable public case study and lifted institutional credibility. Parishes considering an AI pastoral tool can replicate the pattern without replicating the budget by partnering with a local seminary or divinity school. The upfront hours invested in theological review consistently pay back as credibility and operational resilience.
Case Study: Life.Church AI Sermon Preparation at Scale
Life.Church is a United States multi-site evangelical megachurch with roughly 40 campuses, and the ministry publicly disclosed in 2025 that its staff use generative AI during sermon preparation. The move matches Barna’s finding that 41 percent of Protestant pastors do the same across the United States church landscape. The problem Life.Church identified was the volume of content needed to feed roughly 40 campuses and a very large digital following with coherent weekly messaging across all services. The solution deployed is an internal workflow in which AI drafts outlines, cross-references scripture, and surfaces illustrations, which human pastors then edit, correct and preach. Impact has been measurable: faster sermon production without a drop in theological depth, estimated at a 30 percent reduction in draft hours per week. The savings in staff hours let more pastoral attention go to counseling and family visits during each weekly cycle across the campuses. The limitation Life.Church publicly acknowledged is that AI drafts still contain scriptural misattributions, which require careful human verification before any content ever reaches a congregation.
The broader lesson is that AI can be a legitimate productivity tool for ministry when it is kept behind a human review gate. The Life.Church pattern of AI drafting with human finalization matches the design principles emerging from both Catholic and secular research. The pattern exposes the risk that smaller parishes, without a review workflow, may adopt the same tools without the same discipline. Barna’s generational data suggests that younger pastors are more likely to deploy AI without institutional guardrails, which is where most future problems will likely originate. Denominations that provide training and templates to small parishes meaningfully reduce that risk and lift ministry capacity. Life.Church has shared parts of its workflow publicly, which other ministries have begun to adapt.
Case Study: Al-Azhar’s 2024 Statement on AI Muftis
Al-Azhar University in Egypt is the leading Sunni scholarly institution in the global Muslim world. It launched a formal statement in 2024 warning against AI chatbots as sources of religious rulings, in response to proliferating AI mufti apps across the Arabic-speaking web. The core problem was that several startups had rolled out products that would issue a fatwa on demand, which raised serious juridical controversy inside the scholarly community. The solution Al-Azhar proposed was a clear doctrinal line that religious rulings require trained human scholars operating within established methodology, not an AI chatbot in any form. Impact has been visible across the region: several Gulf regulators and the Diyanet in Turkey have cited the Al-Azhar statement in their own guidance to users. The limitation acknowledged in the statement is that AI chatbots still will be used informally in Muslim communities regardless of institutional disapproval or formal rulings. Al-Azhar recommended that Muslim communities invest hours into AI literacy training for scholars and lay users together, in order to raise the baseline understanding.
The case highlights how similar the global religious response to AI has been across traditions. Al-Azhar’s reasoning parallels the Vatican’s Antiqua et Nova almost point for point despite the two institutions sharing no doctrinal framework. Both insist that religious authority is personal, relational and juridically defined, and that AI cannot carry it. Both warn against the vividness of AI output seducing users into treating it as more than a tool. Both call for institutional investment in literacy and formation rather than reactive bans. The convergence itself signals the shape of the global AI governance conversation, which suggests that interfaith cooperation on AI policy is more plausible than it would appear from the outside.
Frequently Asked Questions About the AI Confessional
What is the AI confessional at Deus in Machina?
The AI confessional was an art and research installation inside Peterskapelle in Lucerne, Switzerland. Visitors spoke with an AI Jesus avatar built on OpenAI GPT-4o. The project ran from August 23 to October 20, 2024.
Did an AI really hear Catholic confessions in Switzerland?
No, the AI did not hear a valid Catholic confession. The project team said publicly that the booth was never a sacrament. Catholic teaching requires a validly ordained priest to grant absolution.
Who built the AI Jesus and which model does it use?
The Immersive Realities Research Lab at the Lucerne University of Applied Sciences built it with Catholic theologian Marco Schmid. The underlying language model was OpenAI GPT-4o running on commodity server hardware in Switzerland. The booth supported roughly 100 languages via a translation layer.
How many people used the AI confessional?
The installation drew over 1,000 visitors during its two-month run. The system logged about 900 conversations in total across its two-month run inside Peterskapelle. Of 230 feedback respondents, roughly two thirds described the encounter as spiritual.
Does the Catholic Church allow AI pastoral care?
The Vatican note Antiqua et Nova in 2025 allows AI as a tool for research and catechesis. It explicitly rejects any AI claim to sacramental or juridical authority inside a Catholic liturgical setting. Pope Leo has publicly described AI governance as a moral priority.
What are the main risks of an AI confessional?
Risks include doctrinal drift, pastoral harm to vulnerable users, privacy breaches, and idolatry framing. Even well-designed systems with careful prompts can still produce hallucinated theology under certain conversational conditions. Each risk needs a specific technical and policy safeguard before any parish deploys a pastoral AI tool.
Can an AI chatbot replace a priest?
No AI chatbot can be ordained or carry sacramental authority. A chatbot can help a user prepare for confession or counseling with a human minister. The line between assistant and minister needs to be drawn explicitly in design and disclosure.
How many pastors already use AI in their ministry?
The 2026 Barna and Gloo report found that 41 percent of United States Protestant pastors use AI for Bible study preparation. Nearly two thirds of sermon-preparing pastors use AI in some part of their process. Only 12 percent of Protestant pastors report feeling personally comfortable teaching their congregation about AI topics.
Are there Hindu, Buddhist, Jewish and Muslim equivalents to AI Jesus?
Yes, Hindu developers released GitaGPT with Krishna’s voice drawing on the Bhagavad Gita. Kodaiji temple in Kyoto runs the Mindar robot bodhisattva as a Dharma teaching device. Al-Azhar in Egypt publicly warned against AI muftis in 2024.
What does Antiqua et Nova say about AI in religion?
The 2025 Vatican note frames AI as a product of human intelligence, not an intelligence itself. It affirms AI as a tool for research and administration. It rejects any AI claim to sacramental or juridical authority.
How should a parish evaluate an AI pastoral tool?
Define the task clearly in writing before buying or configuring any AI pastoral tool. Confirm the task does not touch sacraments, juridical rulings, or anything requiring valid ordination. Secure authority approval, review the design against safety principles, and establish a human escalation path for crisis cases.
Is my data from an AI chatbot private the way confession is private?
No, software cannot be bound by the seal of confession. Transcripts can be subpoenaed by courts, logged by vendors indefinitely, or breached in a security incident. Any pastoral AI should minimize retention and tell users clearly how their data is handled.
Can the AI confessional be reopened or copied by other parishes?
Technically yes, since the underlying hardware, model, and prompt components are now commodity items anyone can buy. Doing so safely requires theological supervision, clear labeling, and operational discipline. Copying Deus in Machina without those safeguards is a documented risk.
What is the generational split on AI spiritual advice?
Barna’s 2026 data shows that 40 percent of Gen Z and millennials in the United States trust AI spiritual advice as much as a pastor’s. The overall United States adult share who trust AI spiritual advice as a pastor’s is 30 percent. Only 12 percent of pastors feel equipped to teach on AI, which widens the gap.
Where is the AI confessional conversation headed in 2026 and beyond?
Vendor pricing is pushing AI deployment into even small parishes. Vatican and interfaith statements have drawn a clear line against sacramental impersonation. The near-term outcome depends on how fast denominations publish practical AI policies.
Explore More from AI
How to Spot a Deepfake | How to spot a deepfake in 2026: a 12-sign checklist, a 60-second verification workflow, and honest limits of detection signs.
How is AI Improving Weather Forecasting? | See how AI weather forecasting beats supercomputers, saves billions, and predicts hurricanes faster, plus where the models still fall short today.
FinOps for Agentic AI | Cut surprise AI bills: FinOps for agentic AI covers token budgets, cost per resolved task, attribution, showback, and guardrails for autonomous agents.