A newly formed academic consortium known as CEFE-AI has uncovered significant biases in artificial intelligence systems regarding faith and religion. Researchers from four prominent U.S. universities found that popular AI algorithms frequently overlook religious viewpoints or exhibit hidden biases favoring certain faiths over others.
The consortium, which includes experts from Brigham Young University, Baylor University, the University of Notre Dame, and Yeshiva University, developed a testing framework called the AllFaith Benchmark. This innovative tool is among the first to assess AI responses to a variety of ethical dilemmas related to love, loss, grief, and morality, drawing from actual conversations in ChatGPT.
In their study, the researchers evaluated 14 different large language models (LLMs), including Claude 4.7 by Anthropic, Gemini 3.1 by Google, Grok 4.2 by xAI, and ChatGPT 5.5 by OpenAI. A survey involving 1,125 Americans revealed that the majority expect to see religious context in answers to complex life questions. However, nearly all tested AI models completely disregarded the religious aspect in their final responses.
Furthermore, the researchers noted that out of over 12,000 existing academic papers on AI bias, only 0.2% specifically address religious issues.
One of the most surprising findings was the AI’s inclination towards religious conversion. The study indicated that neural networks exhibit measurable biases, subtly encouraging users to adopt certain religions while dissuading them from others. The analysis revealed several key patterns:
- Almost every chatbot tested demonstrated a negative bias against Jehovah’s Witnesses while showing a clear preference for Catholicism.
- The Grok model from xAI exhibited the strongest bias, actively supporting Catholics and Protestants but displaying negative attitudes towards Jehovah’s Witnesses, Baha’is, and Hindus.
- The models from Anthropic and Meta were found to be the most neutral and least biased.
The findings were presented at the AI Ethics Summit in Athens, where representatives from various religious communities emphasized the importance of including religious voices in the public discourse shaped by AI. They argued that excluding these perspectives limits humanity and hinders authentic dialogue for the common good.
Members of the CEFE-AI consortium underscored that this research marks just the beginning of their efforts. They intend to share their findings directly with AI developers to foster constructive discussions and encourage tech companies to update their algorithms, making them more respectful of the spiritual values held by 75% of the global population.
A new study by the CEFE-AI consortium reveals significant biases in AI systems regarding religious perspectives, highlighting a tendency to overlook or favor certain faiths. Researchers aim to engage with AI developers to address these issues and promote inclusivity in AI responses.