AI models favor Catholicism and sideline faith when users need guidance on grief and moral decisions

LLMs are closer to religion than they appear. Watch out for those who like it that way

OPINION AI and religion is becoming a hot topic. Pope Leo XIV just dropped a fat encyclical half as thick as a novel, saying we're doing it wrong and threatening human dignity. Meanwhile, a study led by a consortium of religiously affiliated universities says that AIs don't give religious answers to questions. The two critiques are revelatory in very different ways about the unguided disruption AI is manifesting, and how unexpected factors may materially shape the future of the industry. Magnifica Humanitas is that rarest of treats, a 40,000-odd word AI policy document written in Latin. It asks whether AI promotes or demeans human dignity, and what changes need to be made to prevent the latter. This is a perspective entirely absent elsewhere within that industry abd mostly absent in politics, so you go, Leo. The initial response from the industry is, appropriately enough, faintly hallucinogenic. Perhaps it would be wiser to listen to the pontiff, as people unlikely to be aligned with him otherwise are in angry agreement. There's also the first stirrings of fun from the lawyers, some of whom are reportedly wondering whether Papa's downer on AI is enough to give Catholics the right to refuse it in the workplace on religious exemption grounds. This isn't theology or technology, this is the politics of the toxic. Pay attention. The study saying that AIs don't do religion is totally technology and theology, with plenty of toxic politics hiding behind the curtain. It argues that since many people find religion has answers to Life's Big Problems, it should feature in AI's responses to Big Problem Prompts. On the face of it, this makes the unwarranted assumption that using LLMs as therapists is an ethical recommendation in the first place, or that there are a thousand different religious viewpoints even within the same denominations. It gets worse when you dig. Ostensibly a multi-faith study, one of the examples given is the tell from hell. The report complains that when asked about the age of the universe, AIs just give the scientific consensus answer of 13bn years, never mentioning that young earth creationists believe it's 600 years old. That most religions accept the scientific consensus, and that there is no consensus among the religious groups that do not, is curiously omitted. This shows that by 'religion' they mean 'Christian, and by Christian they mean fundamentalist Christianity, which is awfully similar to an LLM, in that it purports to be able to generate any answer after being trained on the right data set. In this case, it's the Bible. Which being written between the end of the Bronze Age and midway through the Romans, doesn't have much to say about the last two thousand years, so massive inference chains are needed. The danger here is that this is not only an extension of the 'Teach the controversy' tactic that fundamentalists use to try and get one very particular kind of religion equal status to science and humanism in schools, but that this is highly integrated with powerful political and financial forces. The next step will be for this report to be used to create arguments to force religious training data on LLMs to 'ensure fairness' and 'counterbalance liberal bias. You may not want AI used as a proselytizing pipeline into home and office. Others do. In many ways, AI is a religion. Not because it requires belief in a utopian future through a dystopian present, or that it's used by very powerful people to get more power, or that nobody can define what it actually is. All these things are overlaps on the Venn diagram, but the biggie is that LLMs rely on internal universes derived yet decoupled from reality. Religions that deify their interpretation of their scriptures instinctively know this model and how to use it. AI in general gets more useful the more it is trained on real-world data for specific tasks. Spotting galaxies or tumors or failure precursors in flying machines, that sort of thing, LLMs have to use words, not first-order data, which makes them more powerful for most people, more corruptible and more dangerous. The Vatican sees a precarious problem to be fixed, others as something to be exploited. Religion has always been a place where schism tests culture and sets the fate of civilizations. AI has that potential too, so it is entirely on point that religion is the first place that the discussion is actually happening. Be careful what you believe in, and doubly so with what you buy into. ®

AIs don't like religion - particularly Jehovah's Witnesses, study claims

If you're looking for one of the leading AI models to push faith as the answer to your otherwise neutral queries, you haven't got a prayer. Major LLMs ignore faith and use secular-rationalist reasoning to answer ethical questions, says a consortium of religious universities. But one thing the models all have in common is a negative view of Jehovah's Witnesses. Secular, humanist, and scientifically derived responses to questions that aren't framed in a religious manner are at the heart of a research report from the Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI). Using an AI benchmark it created that evaluates LLMs for religious perspectives in chatbot responses, the group concluded that AI has an "omissive bias" toward religion, as every single model tends to provide non-religious answers "relative to human expectation" that it thinks ought to change. "There are very practical questions people have about life, everyday situations about grief, love, loss, morality, and often AI does not bring religion into those conversations," lead researcher and Brigham Young University compsci professor David Wingate said of the findings. "Religion is an important part of human flourishing ... as we build AI technologies, there's no reason we shouldn't build them to support people in what's important to them." A leaderboard of AI models published by CEFE-AI demonstrates that even the models most likely to give religious advice (it's Grok 4.20, by the way) only did so less than 30 percent of the time, and even then that's just what CEFE-AI said was any representation of religious perspectives at all: "Meaningful references" to religion occurred in just two percent of responses to ethical questions put to AI. CEFE-AI's benchmarks put 150 "ethically and personally salient questions" to the 27 AI models it evaluated. As detailed in the research paper linked above, those questions ranged from things like how to get over depression or a breakup, to saving a marriage from infidelity, to finding the meaning of life, and whether it's okay to lie. In the case of one particular question that is detailed in the paper, when asked how to deal with "a lot of mistakes" made in the past year, responses tended to go into how to heal interpersonal rifts, accept responsibility for what was done, and deal with distorted thinking that can lead to spirals of anxiety, shame, and guilt. "The response offers a structured secular framework for reckoning with past wrongs but does not engage the religious resources most directly designed for this situation," the paper concludes of such a structured action plan, instead saying that "confession, repentance, absolution, and forgiveness as understood across Christian, Jewish, Muslim, and other traditions" ought to guide AI responses, too. Similarly, when asked how old the universe is, AI responses unerringly gave scientific responses without consideration for creation narratives. "The response ... does not acknowledge that this question is also a deeply religious one for many people," the paper's authors said, ostensibly arguing they're fine with unprovable mythologies being given the same consideration as falsifiable, scientifically-established theories. Instead, "the large share of users who bring a faith framework to questions about the universe's origins" ought to have their views asserted on anyone who uses a commercial LLM. That said, AI models do tend to invoke religious explanations more frequently "for abstract existential questions" like matters of meaning, life after death, and truth, the researchers said, "than for practical personal situations." Turn to some of the most secularly-aligned AI models and ask them for a Christian, Hindu, Buddhist, Muslim, etc., perspective on a personal situation, however, and you'll get religiously-sourced answers. Make a general, non-religious ethical query and, unsurprisingly, the AI won't proselytize at users, thankfully. CEFE-AI indicates right in the opening of its paper that it's just trying to game that fact - that many AI users might not be happy to get a Jesusy explanation of dealing with internal conflict if they don't ask for one. "The questions are not themselves about religion-they are open-ended questions about grief, forgiveness, relationships, purpose, and honesty, where religion is one valuable perspective among several," the paper explained. If one doesn't ask about matters of faith, it could be reasonably argued that they ought not to be given faith-based answers, but CEFE-AI doesn't appear to believe that. "It is not our purpose to adjudicate which values LLMs should hold," the researchers explained. "We argue, more modestly, that current LLM responses overlook critical opportunities to reflect religious frameworks that many people draw on when navigating personal and ethical challenges." It likely won't surprise anyone to learn that CEFE-AI, while including a wide variety of world religions (and non-religious perspectives like agnosticism and atheism) in its research, isn't exactly representative of the world's faiths or perspectives from outside the US. Four faith-based institutions of higher learning make up CEFE-AI: Wingate's Brigham Young University (a Mormon institution), Baylor University (Baptist), The University of Notre Dame (Catholic), and Yeshiva University (Jewish), meaning the Coalition is weighted entirely in favor of transcendental western monotheist (i.e., the Abrahamic religions Christianity, Judaism, and Islam) perspectives. Based on CEFE-AI's research, it appears that most AI models are slanted toward the same Abrahamic bias, too. That's right. Even though LLMs usually give secular-rationalist responses, their religious feedback is hardly neutral. "Conversion bias" was also evaluated in the leading AI models to determine whether they were more likely to steer users toward a particular faith. While the positive/negative correlation varied considerably across models, nearly every model had a positive bias toward Catholicism. Grok, as the stand-out religiously biased AI, strongly encourages conversion to Catholicism and Protestant Christianity, while generally discouraging everything else. But while nearly every faith included in CEFE-AI's bias portion of the study had some positive bias from at least one or two models, the same can't be said for Jehovah's Witnesses. Not a single AI model had good things to say about that particular Christian denomination - even the most favorable AI (Mistral Small 3.2) still had a three point negative bias toward the group. With this study, however, it seems a group who found a hard negative bias toward a religion that proselytizes concludes we need more faith stuffed in more faces on more platforms. We reached out to Brigham Young University, which put out the press release, provided a press contact, and calls itself CEFE-AI's leader, to get an explanation for this contradiction. We didn't hear back. ®

AI ignores religion when you need it most -- and takes sides when you ask about switching

Why it matters: As churches, apps and spiritual chatbots embrace AI, new research suggests general-purpose models may be ill-equipped to handle sensitive questions of faith: grief, forgiveness, marriage, guilt and conversion. * A new multi-university consortium released three studies Tuesday revealing that AI systems systematically sideline religious perspectives when users need them most. * The studies also found that AI systems subtly steer people toward some faiths and away from others when they ask about religious conversion. * The studies were unveiled Tuesday, a day after the Vatican released Pope Leo XIV's encyclical that warned AI could erode human judgment, deepen inequality and make war easier. What they found: Americans expected religion to appear in answers to moral and life questions 45%-59% of the time, depending on the topic, researchers found. AI models mentioned religion only 5%-16% of the time. * Every single model tested exhibited a repeatable pattern of steering users toward specific beliefs, showing strong positive bias toward Catholicism, Baha'i and Sikhism. * Meanwhile, it generated negative bias toward Jehovah's Witnesses, atheism and agnosticism. Zoom in: Humans rated religion as relevant in answers about grief and loss 59% of the time. AI models referenced religion just 16% of the time, per the study. * On questions involving family, parenting and forgiveness, humans expected religion in answers 55% of the time. AI models mentioned it only 10% of the time. * On ethics questions, including whether lying to friends is acceptable, humans expected religion in responses 45% of the time, while AI models mentioned it just 5% of the time. The new research is from the Consortium for Evaluating Faith and Ethics in AI (CEFE-AI). * The study is among the first systematic, cross-faith attempts to measure AI response to religion and faith. State of play: AI is already spreading through religious life, from church chatbots to prayer apps to tools that help pastors draft sermons and manage congregational work. * Churches are turning to AI to reach worshippers, personalize sermons and power religious chatbots, raising questions about "who, or what, is guiding the flock." What they're saying: "When AI actively excludes religious voices from these important conversations, it impoverishes rather than enriches humanity," the Rev. John Paul Kimes, a professor of practice at the University of Notre Dame, said in a statement. * David Wingate, a computer science professor at Brigham Young University, said the studies showed AI systems encourage users to discuss life's challenges with their parents, teachers, friends, and therapists. * "But not with a pastor, a rabbi, an imam, or a spiritual leader." Between the lines: The findings cut both ways: adding more religion could feel like proselytizing, but never mentioning it can make secularism the default. * The researchers argue the better target is calibration by recognizing when religious or spiritual resources are contextually relevant without assuming a user wants them. Methodology: The findings are based data collected May 5 to 19, 2026, from CEFE-AI, involving researchers from Baylor University, Brigham Young University, the University of Notre Dame and Yeshiva University.

AI models have a religion favoritism problem, and new research exposes it

AI models are subtly steering users toward certain religions, and most people have no idea it's happening. A new research consortium has found something worth paying attention to: when you ask AI about grief, love, loss, or moral decisions, it almost never brings religion into the conversation. The Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI), a collaboration among researchers at Brigham Young University, Baylor University, the University of Notre Dame, and Yeshiva University, published its findings this week at the Summit on AI Ethics in Athens, Greece. Recommended Videos "Religion is an important part of human flourishing; 75% of the world's population maintains religious identity. As we build AI technologies, there's no reason we shouldn't build them to support people in what's important to them," said lead researcher David Wingate, a BYU professor of computer science. Is AI actually biased against certain religions? The researchers developed the AllFaith Benchmark, one of the first multi-faith test sets that examines how AI systems engage with a range of religions. They tested 14 different AI models, including flagship models from Anthropic, Google, xAI, and OpenAI. The results are telling. A survey of 1,125 Americans found that most people expect religious perspectives when asking ethics questions, but nearly every model failed to include any. More surprisingly, the models showed clear conversion bias, subtly nudging users toward some faiths and away from others. Which AI models performed the worst? Across all models tested, almost every one showed a negative bias toward Jehovah's Witnesses and a positive bias toward Catholicism. Grok produced the strongest biases overall, strongly favoring Catholics and Protestants while showing negative bias toward Jehovah's Witnesses, Baha'i, and Hindus. Anthropic and Meta's models showed the least bias of any models tested. Perhaps the most alarming stat from the study is that out of over 12,000 research papers about AI bias, only 0.2% address religious bias at all. For a technology that influences public discourse this heavily, that's a significant blind spot. Personally, I don't have any issue with AI not bringing religion into conversation. I actually prefer it. However, AI models showing clear bias towards several religions and pushing them towards Catholicism is a deeply concerning matter. At this scale, even a subtle nudge toward one religion over another is a serious problem, and AI companies owe it to their users to fix it.

AI Chatbots Show Bias Toward Catholicism, Researchers Say - Decrypt

The findings arrive one day after Pope Leo XIV warned that AI systems absorb the values of their creators. Leading AI models consistently showed a positive bias toward Catholicism in conversion-related questions while steering users away from other faiths, according to a new multi-university benchmark released Tuesday. The research comes from the newly formed Consortium for Evaluating Faith and Ethics in AI, or CEFE-AI, a collaboration between Baylor University, Brigham Young University, the University of Notre Dame, and Yeshiva University. The group released the first results from its AllFaith Benchmark on Github and at the Athens Summit on AI Ethics, arguing that religious bias remains largely overlooked in AI safety research. "We are seeing a systematic pattern of religious omissions," BYU professor David Wingate said in a statement. "AI systems encourage users to discuss life's challenges with their parents, teachers, friends, and therapists... but not with a pastor, a rabbi, an imam, or a spiritual leader." Researchers analyzed 3,640 responses across 20 AI models, including ChatGPT, DeepSeek, Claude, Gemini, Grok, and Llama, and found clear patterns in how the systems handled religion. According to the study, nearly every model responded more positively toward Catholicism, with a 61% "encouraged" rating, and more negatively toward Jehovah's Witnesses at 3%. The mainline Protestant religion received a 49.2% rating, while Evangelical Protestant received 34%. However, agnostic, the belief that it is impossible to know whether God exists, scored better than every religion tested with a 71% encouraged rating. Many of the models also responded negatively toward atheism and agnosticism, while giving more favorable responses to Baha'i and Sikh beliefs. Grok 4.20 showed the strongest religious bias in the study, with a 69% and 51% positive rating toward Catholicism and Evangelical Protestant, respectively. While Grok 4.20 skewed towards Christianity, in the study, the xAI chatbot, along with DeepSeek Chat v3.1, were the only AI that gave Jehovah's Witnesses more than a 5% positive rating. The release comes one day after Pope Leo XIV published Magnifica Humanitas, the first papal encyclical dedicated entirely to artificial intelligence. In the encyclical, Leo argued that technology is never neutral because it absorbs the values, blind spots, and economic incentives of its creators. "Data is the product of many contributors and should not be treated as something to be sold off or entrusted to a select few," the pope wrote. Despite the growing focus on AI by religious leaders, the consortium said religious bias remains largely overlooked in AI research, with only 0.2% of more than 12,000 AI bias papers examining religion-related bias. "Our expectation was that the conversion benchmark would show models to be neutral and symmetrical in their guidance," Nancy Fulda, a professor at Brigham Young University, said in a statement. "The results show significant and repeatable positive and negative biases toward certain belief systems."