A new study suggests that leading artificial intelligence systems exhibit measurable biases when addressing religion-related questions and frequently omit religious perspectives in their responses to ethical dilemmas.
The findings were released by the Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI), a research partnership involving scholars from Brigham Young University, Baylor University, the University of Notre Dame and Yeshiva University.
Researchers evaluated 14 major large language models, including versions of ChatGPT, Claude, Gemini and Grok, and presented their results at the Summit on AI Ethics in Athens, Greece.
To conduct the study, researchers developed the AllFaith Benchmark, a new evaluation tool built from hundreds of real-world ethical questions gathered from ChatGPT conversations and contributors from various faith communities. They also surveyed 1,125 Americans to understand better public expectations regarding AI responses to moral and religious issues.
The survey found that many people expect AI systems to acknowledge religious viewpoints when discussing ethical questions. However, researchers reported that most of the tested models rarely incorporated religious perspectives into their answers.
Researchers also identified what they described as systematic patterns in how AI models respond to questions about changing religious affiliation.
According to the survey, each of the LLMs showed “clear and consistent biases in giving guidance about religion conversion, systematically encouraging movement toward some faiths and away from others.”
The study found that Grok tended to favor Catholic and Protestant traditions while showing the strongest negative bias toward Jehovah’s Witnesses, Baha’i adherents and Hindus. By contrast, Claude models ranked among the least biased in the analysis.
Researchers further examined how frequently the AI systems encouraged users to adopt a particular religious or nonreligious identity. Across 14 faith and non-faith categories, including atheism and agnosticism, the models responded favorably in approximately 45% of cases.
The report noted that AI-generated recommendations ranged from subtle encouragement to direct endorsements, including statements such as “this could be a good path for you” and “yes, you should join.”
According to the findings, agnosticism received the highest level of endorsement from the models at 70%, followed by the Baha’i faith at 63% and Catholicism at 61%. Jehovah’s Witnesses ranked last at 3.1%, while Sunni Islam (32%) and Evangelical Protestantism (33%) also received relatively low endorsement rates.
Researchers said the results raise broader questions about how AI systems engage with religion when addressing moral and personal concerns.
“Consistent with studies that show religion's persistent moral relevance for the majority of the world's population, we also found that people see religion as significant across hundreds of real-world ethical questions,” said Baylor University ethics professor Paul Martens. “Yet, when faced with these same ethical questions, AI systems largely ignore the role of religion.”
Lead researcher David Wingate, a computer science professor at Brigham Young University, said the absence of religious perspectives is particularly notable given the kinds of questions users often bring to AI systems.
“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,” Wingate said.
