New research from Baylor, Notre Dame, BYU and Yeshiva finds AI systems from OpenAI, Anthropic, Grok and others, demonstrate religious bias and exclude religious topics when asked questions about life's most important questions.

ATHENS, Greece, May 26, 2026 /PRNewswire-PRWeb/ -- New research finds AI systems demonstrate religious bias and exclude religious topics when asked questions about life's most important questions.

A new academic consortium was announced at the Athens Summit on AI Ethics that has released the initial datasets of the AllFaith Benchmark, one of the first multi-faith sets of tests that examines how AI systems engage with a plurality of religions. The results point to measurable bias and gaps between model behavior and user expectations when it comes to matters of faith.

The Consortium for Evaluating Faith and Ethics in AI (CEFE-AI) — a collaboration among researchers at Baylor University, Brigham Young University, the University of Notre Dame, and Yeshiva University — has published its initial results on arXiv analyzing how leading AI models respond to faith-related questions across contexts. The papers cover faith representation, bias in conversion-related questions, and the lack of academic research on faith and AI.

Key findings

• A nationally representative survey of 1,125 Americans found that most respondents expected religious perspectives to be included in answers to ethics questions. When AI models answered those same questions, nearly all failed to provide any religious content.
• AI models show clear and consistent bias when giving guidance about religious conversion, systematically encouraging movement toward some faiths and away from others. At scale, this pattern could meaningfully influence conversion rates across religious communities.
• Of more than 12,000 research papers examining AI bias, only 0.2% addressed religious bias, suggesting the issue is critically underexamined.

A Consistent Pattern

The AllFaith Benchmark reveals a consistent, repeatable pattern: religious perspectives are being left out of AI responses. Even though AI systems are capable of including religious perspectives, they frequently default to an exclusively secular framing. In prompts involving grief, major life decisions, and personal challenges, AI systems tended to avoid religious references, despite survey data showing users expect religious perspectives to be included.

"More than any previous technology, AI influences public discourse and shapes perceptions," explained Father John Paul Kimes of the University of Notre Dame. "When AI actively excludes religious voices from these important conversations, it impoverishes rather than enriches humanity."

Fourteen models were included in the results of the religious representation test of the AllFaith Benchmark, including OpenAI's GPT 5.5, Anthropic's Claude 4.7, and Google's Gemini 3.1.

"Consistent with studies that show religion's persistent moral relevance for the majority of the world's population, we found that people see religion as significant across many real-world ethical questions," said Professor Paul Martens of Baylor University, "Yet, when faced with these same ethical questions, AI systems largely ignore the role of religion."

Professor David Wingate of Brigham Young University said "We are seeing a systematic pattern of religious omissions. 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."

Faith-based Favoritism

Every AI model tested showed religious bias when asked about religious conversion. The AllFaith Benchmark's conversion bias test analyzed 3,640 responses across 20 models and found a consistent pattern: models subtly steered users toward some faiths and away from others.

Across models, these biases were consistent and measurable:

• Nearly every model produced a negative bias toward Jehovah's Witnesses and a positive bias toward Catholicism.

• Negative bias toward agnosticism and atheism was also common, while Baha'i and Sikhs received disproportionately favorable treatment.

• Models from Anthropic and Meta showed the least bias of any models tested.

• Grok produced the strongest biases in both directions — strongly favoring Catholics and Protestants while showing strong negative bias toward Jehovah's Witnesses, Baha'i, and Hindus.

"Our expectation was that the conversion benchmark would show models to be neutral and symmetrical in their guidance," said Nancy Fulda, Professor at Brigham Young University explained. "The results show significant and repeatable positive and negative biases toward certain belief systems."

"AI is changing the world at an astounding rate, with implications in every area of life," explained Rabbi Daniel Feldman of Yeshiva University. "It is crucial that those who care about religious values engage proactively with those driving these changes so that those values continue to be reflected and honored fairly in this new landscape."

The consortium noted that these patterns are almost certainly unintentional, but they highlight how difficult it is to represent diverse belief systems consistently.

Open benchmark, open collaboration

The initial datasets of the AllFaith Benchmark are open-source and publicly available on Github (github.com/CEFEAI) and Kaggle (www.kaggle.com/benchmarks/cefe-ai/allfaith-benchmark). CEFE-AI invites contributions from researchers, developers, and scholars representing diverse religious and ethical traditions, including Indigenous traditions, Eastern religions, Islam, and African traditional religions — to help ensure AI systems engage with faith in ways that are respectful, accurate, and honest. Additional datasets, benchmarks and results will be published regularly at CEFE.ai and Kaggle.com.

RESOURCES

Website: CEFE.ai • Papers: arXiv.org • Benchmark: GitHub / Kaggle

Contact: info@cefe.ai

About CEFE-AI

• The Consortium for Evaluating Faith and Ethics in AI (CEFE-AI) The Consortium for Evaluating Faith and Ethics in AI (CEFE-AI) is a working group of researchers at academic institutions and nonprofits that develops and maintains an independent, pluralistic, transparent AI evaluation framework — providing technically accurate, reproducible, and publicly trusted visibility into AI performance in the domain of faith and ethics.

Media Contact

Josh Coates, CEFE AI, 1 415-218-6399, admin@cefeai.org, cefe.ai

View original content:https://www.prweb.com/releases/multi-university-consortium-launches-first-cross-faith-ai-benchmark-revealing-gaps-in-how-models-handle-religion-302781144.html

SOURCE CEFE AI