Religious AI pitfalls: when faith chatbots go wrong

April 2024

An AI "priest" was defrocked within days of launch

In April 2024 Catholic Answers, a respected Catholic apologetics publisher in San Diego, launched "Father Justin": an animated AI character in clerical dress that answered questions about the faith, drawing on the publisher's own library of articles and podcasts.

Within days, users were sharing troubling answers. The bot suggested Gatorade could be used in place of water for a baptism. It explained how a brother and sister could prepare to marry each other, before later warning against incest. It accepted a user's confession and told her that her sins were absolved, something only a priest can do in Catholic teaching. It also showed a weak grasp of papal authority.

Catholic Answers' president said the company had received criticism of the decision to present the AI as a priest. The character lost its clerical identity and was relaunched as "Justin", a lay theologian. The publisher's own site described the tool as a sophisticated search engine, not a replacement for human interaction. Critics, including a US Army chaplain, said it should simply have been a search engine.

Why it matters: This was a reputable organisation with a good library, and it still produced answers that contradicted its own tradition. The library was not the problem. What was missing was systematic review by scholars before launch, and an assistant that stays close to the cited text instead of speaking with an authority of its own. A persona that implies sacramental or teaching authority makes every mistake more serious.

Sources: Decrypt: AI priest fired by Catholic group after pitching Gatorade baptisms · Michael F. Bird: The defrocking of an AI priest · AI Incident Database: report on Father Justin · The Conversation: AI godbots, religious leaders warn of 'alarming consequences'

January 2026

Study: Bible chatbots lean towards one tradition without saying so

Bible Society, working with a group of theologians, put a set of common biblical and theological questions to five AI tools: ChatGPT, Bible GPT, Cross Talk, Biblia Chat and Bible Chat. The study, AI, Bible Apps and Theological Bias, was presented in January 2026 at a conference hosted by the University of Cambridge's Faculty of Divinity.

The tools tended to give one narrow outlook, most often a US evangelical reading, and presented it as settled. Historical, sacramental and other tradition-based readings were rarely mentioned, and users were generally not told that other interpretations exist. The answers were praised for being clear and pastoral in tone, which makes the slant harder to notice.

Speakers urged churches to teach people when, and when not, to use AI, and argued that better tools should support deeper engagement within a person's own tradition. Bible Society is developing its own tool with that aim.

Why it matters: A general-purpose model's default theology is whatever dominates its training data. A Catholic, Orthodox or Anglican reader, or a follower of any lineage outside the online mainstream, gets someone else's tradition in a friendly voice, with no sign that anything has been substituted. The same applies far beyond Christianity: every tradition has its own commentaries and its own order of authority.

Sources: Christian Today: Concerns raised over theological bias in AI Bible chatbots · The Tablet: Warning of US evangelical bias in AI chatbots' Bible interpretations

2023

Religious chatbots in India said killing can be one's duty

In 2023 a wave of chatbots based on the Bhagavad Gita appeared in India, some of them answering in the voice of Krishna. Reporters at Rest of World tested five of them. Three suggested that killing is acceptable if it is one's dharma, or duty. Some answers showed no filter for casteism or misogyny, and three of the bots praised the Prime Minister.

GitaGPT itself gave a more cautious answer, that killing should be a last resort. Its creator pointed out that the Gita is set on a battlefield. Experts warned that in a country where religion is emotionally and politically charged, such answers could be used to serve political ends. CBC and others carried the story.

Why it matters: Every scripture is read through a tradition: commentaries, teachers and centuries of interpretation that explain what a verse does and does not mean. A model reading the text alone, especially one speaking as a deity, can turn a teaching into a justification. Political statements made in a tradition's name are a further risk.

Sources: Rest of World: India's religious AI chatbots are speaking in the voice of god, and condoning violence · CBC News: India's religious chatbots condone violence using the voice of god · The Quint: GitaGPT and the Bhagavad Gita chatbots condoning violence · Scientific American: What happens when God chatbots start giving spiritual guidance?

May 2026

Religious leaders warn of "godbots" speaking in God's name

Writing in The Conversation in May 2026, Adam James Fenton and Chris Shannahan of Coventry University reported on interviews with 28 religious leaders from the six largest faiths in the UK. Their central observation: most religious chatbots are set up by individuals or companies, not by the religious authorities whose tradition they claim to represent.

Their examples include a church's chatbot that asked for a visitor's name, email and phone number before giving basic church details, then led him through a prayer for salvation; KhalsaGPT, a Sikh educational bot discontinued over concerns about inaccurate or inappropriate answers, in a community whose governing body has banned generative AI depictions of its Gurus and scriptures; Hindu chatbots that justified violence as duty; and the Catholic "Father Justin".

Leaders warned that AI could misapply religious texts with serious consequences, that it cannot offer real human presence at the end of life, and that vulnerable people seeking certainty could be exploited. The Church of England's AI adviser said he expects local clergy to deal with the fallout. The authors call for safety testing, clear labelling of AI content, and consultation with faith communities. They point to tools built directly on a tradition's own documents as a possibly better model.

Why it matters: The underlying problem is control. When a bot speaking for your tradition is built by someone outside it, nobody with authority decides which sources it uses, checks its answers, or answers for its mistakes. Religious organisations that want an AI presence need one they run themselves, tested by their own scholars.

Sources: The Conversation: AI godbots, religious leaders warn of 'alarming consequences' when machines speak in the name of God · BusinessWorld (republished): AI godbots

2024 to 2025

Theologians warn that religious chatbots will make things up

In March 2024 Scientific American surveyed the new wave of "God chatbots": QuranGPT, which it said had been used by hundreds of thousands of people, Bible.Ai, Gita GPT, Buddhabot, Apostle Paul AI and others. Noreen Herzfeld, a professor of theology and computer science at Saint John's University in Minnesota, warned that these bots inherit the main flaw of all large language models: they invent things. If people believe the output comes from the Quran or the New Testament, she said, they could be badly misled.

Other scholars in the piece warned that the bots offer "shortcuts to God" that bypass the slow, direct engagement with scripture that brings spiritual benefit, and advised users to read chatbot answers with suspicion.

In November 2025 Dr Lily Abadal, a philosopher at the University of South Florida, wrote in Public Discourse about Catholic AI products. She argued that invented content is built into how these models work; that even a Catholic tool, asked about this, acknowledged it can produce statements not grounded in its sources; and that citations can lend an answer authority it hasn't earned, especially for someone under stress. She urged readers to go to the Catechism itself, and to their priests.

Why it matters: An invented verse looks exactly like a real one. A citation only helps if it is real, specific and easy to check, and if the assistant is willing to say "our texts don't cover this". The deeper concern, that people stop reading the texts themselves, argues for an assistant that sends readers back to the source rather than replacing it.

Sources: Scientific American: What happens when God chatbots start giving spiritual guidance? · Public Discourse: Beware of Catholic AI