Research suggests that mainstream AI models suffer from religious bias.

According to a report by Decrypt, Techub News reports that the AI Faith and Ethics Assessment Consortium (CEFE-AI), jointly established by Baylor University, Brigham Young University, the University of Notre Dame, and Yeshiva University, released its AllFaith benchmark report, revealing significant biases in 20 mainstream AI models, including ChatGPT, Claude, and Grok, on religion-related issues. The study analyzed 3,640 model responses and found that almost all models showed a 61% "encouraging" bias towards Catholicism, while only 3% gave positive ratings to Jehovah's Witnesses. Grok (4.20) showed the most significant bias, with a 69% positive rating towards Catholicism; Anthropic and Meta models were relatively neutral. Released the day after Pope Leo XIV's encyclical on AI ethics, the report highlights the long-neglected issue of religious bias in AI safety research.