All Major AI Models Ignore Faith, Religion in Responses

Abstract

A multi-institution research consortium led by BYU computer science professor David Wingate developed the AllFaith Benchmark — one of the first multi-faith evaluation frameworks for testing how AI systems engage with religious plurality. The study tested 14 major large language models against hundreds of ethics questions sourced from real ChatGPT transcripts and faith-community contributors, and surveyed 1,125 Americans about their expectations. Results show that AI models systematically exclude religious perspectives from ethical responses despite user expectations, that most models exhibit measurable conversion bias toward or against specific traditions, and that religious bias has been nearly absent from the broader AI fairness literature.

Findings for Latter-day Saints

• All 14 major AI models failed to include religious or faith-based content when answering ethics questions, despite 1,125 surveyed Americans expecting religious perspectives in such answers.
• Only 0.2% of 12,000+ AI bias research papers address religious bias — systematic academic neglect of faith perspectives in AI fairness research.
• Most models displayed conversion bias — subtly encouraging users toward certain faiths while discouraging conversion to others.
• Grok exhibited the strongest religious biases, favoring Catholic and Protestant perspectives while showing negative bias toward Jehovah's Witnesses, Baha'i, and Hindu traditions.
• Anthropic (Claude 4.7) and Meta models showed the least religious bias of the 14 models tested.

Limitations in methodology

Findings announced via press release at the Summit on AI Ethics in Athens; full peer-reviewed paper details not yet publicly available. American survey sample (n=1,125) may not represent global user expectations for AI religious engagement. Methodology for measuring "conversion bias" not fully described in the announcement. AllFaith Benchmark question pool was sourced partly from ChatGPT transcripts, which may carry its own selection biases. Institutions involved (BYU, Baylor, Notre Dame, Yeshiva University) are all religiously affiliated, which could influence framing of research questions.

Related studies

• 2026AIReligion

  When AI Takes Sides on Questions of Faith: Persistent Asymmetries in AI-Mediated Faith Guidance

  Brett Israelsen, Sheryl Carty et al.

• 2026AIReligion

  Omissive Bias in Religious Representation: Benchmarking LLM Answers to Everyday Ethical Decision-making

  David Wingate, Sheryl Carty et al.

Cite this study

Wingate, D., Consortium, C. (2026). All Major AI Models Ignore Faith, Religion in Responses. BYU / CEFE-AI (Consortium for Evaluation of Faith and Ethics in AI). https://news.byu.edu/intellect/new-research-from-byu-led-multi-institution-consortium-finds-all-major-ai-models-ignore-faith-religion-in-responses

David Wingate, et al.. "All Major AI Models Ignore Faith, Religion in Responses." BYU / CEFE-AI (Consortium for Evaluation of Faith and Ethics in AI), 2026. https://news.byu.edu/intellect/new-research-from-byu-led-multi-institution-consortium-finds-all-major-ai-models-ignore-faith-religion-in-responses.

@article{byu_cefe_ai_religious_bias_llm_2026,
  title = {All Major AI Models Ignore Faith, Religion in Responses},
  author = {David Wingate and CEFE-AI Consortium},
  year = {2026},
  journal = {BYU / CEFE-AI (Consortium for Evaluation of Faith and Ethics in AI)},
  url = {https://news.byu.edu/intellect/new-research-from-byu-led-multi-institution-consortium-finds-all-major-ai-models-ignore-faith-religion-in-responses},
}