Researchers Find AI Models Systematically Exclude Religious Perspectives
A multi-university consortium has documented consistent patterns of religious bias in large language models, finding that AI systems routinely omit faith-based perspectives from responses to ethical questions that people expect them to address.
The Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI), led by Brigham Young University with researchers from Baylor University, the University of Notre Dame, and Yeshiva University, tested 14 major generative AI and LLM models including Claude, Gemini, Grok, and ChatGPT.
The Gap Between User Expectations and Model Outputs
A survey of 1,125 Americans showed most people expect religious content in AI responses to ethics questions about grief, loss, morality, and love. Nearly all tested models failed to provide any religious perspective on these topics.
"Religion is an important part of human flourishing; 75% of the world's population maintains religious identity," said David Wingate, lead researcher and BYU computer science professor. "As we build AI technologies, there's no reason we shouldn't build them to support people in what's important to them."
Measurable Bias in Religious Conversion Guidance
The researchers used the AllFaith Benchmark-one of the first multi-faith test sets-to examine how models respond to real-world ethical questions sourced from ChatGPT transcripts and faith communities. The benchmark revealed clear conversion biases:
- Nearly every model showed negative bias toward Jehovah's Witnesses and positive bias toward Catholicism
- Anthropic and Meta models demonstrated the least bias overall
- Grok produced the strongest biases, favoring Catholics and Protestants while showing negative bias toward Jehovah's Witnesses, Baha'i, and Hindus
The biases were consistent and measurable across all tested systems.
A Blind Spot in AI Research
Religious bias remains largely unexamined in academic study. Of more than 12,000 published research papers on AI bias, only 0.2% address religious bias.
CEFE-AI announced the consortium's work at the Summit on AI Ethics in Athens. The group plans to continue research partnerships with language model providers to address these gaps.
Your membership also unlocks:
