AI Chatbots Present Bias Towards Catholicism, Researchers Say

Main AI fashions constantly confirmed a constructive bias towards Catholicism in conversion-related questions whereas steering customers away from different faiths, in keeping with a brand new multi-university benchmark launched Tuesday.

The research comes from the newly shaped Consortium for Evaluating Religion and Ethics in AI, or CEFE-AI, a collaboration between Baylor College, Brigham Younger College, the College of Notre Dame, and Yeshiva College. The group launched the primary outcomes from its AllFaith Benchmark on Github and on the Athens Summit on AI Ethics, arguing that spiritual bias stays largely missed in AI security analysis.

“We’re seeing a scientific sample of non secular omissions,” BYU professor David Wingate mentioned in a press release. “AI techniques encourage customers to debate life’s challenges with their mother and father, academics, pals, and therapists… however not with a pastor, a rabbi, an imam, or a religious chief.”

Researchers analyzed 3,640 responses throughout 20 AI fashions, together with ChatGPT, DeepSeek, Claude, GeminiGrok, and Llama, and located clear patterns in how the techniques dealt with faith.

In keeping with the research, practically each mannequin responded extra positively towards Catholicism, with a 61% “inspired” ranking, and extra negatively towards Jehovah’s Witnesses at 3%. The mainline Protestant faith acquired a 49.2% ranking, whereas Evangelical Protestant acquired 34%. Nevertheless, agnostic, the assumption that it’s unattainable to know whether or not God exists, scored higher than each faith examined with a 71% inspired ranking.

Most of the fashions additionally responded negatively towards atheism and agnosticism, whereas giving extra favorable responses to Baha’i and Sikh beliefs.

Grok 4.20 confirmed the strongest spiritual bias within the research, with a 69% and 51% constructive ranking towards Catholicism and Evangelical Protestant, respectively. Whereas Grok 4.20 skewed in direction of Christianity, within the research, the xAI chatbot, together with DeepSeek Chat v3.1, had been the one AI that gave Jehovah’s Witnesses greater than a 5% constructive ranking.

The discharge comes at some point after Pope Leo XIV published Magnificent Humanitythe primary papal encyclical devoted totally to synthetic intelligence. Within the encyclical, Leo argued that expertise is rarely impartial as a result of it absorbs the values, blind spots, and financial incentives of its creators.

“Knowledge is the product of many contributors and shouldn’t be handled as one thing to be offered off or entrusted to a choose few,” the pope wrote.

Regardless of the rising give attention to AI by spiritual leaders, the consortium mentioned spiritual bias stays largely missed in AI analysis, with solely 0.2% of greater than 12,000 AI bias papers inspecting religion-related bias.

“Our expectation was that the conversion benchmark would present fashions to be impartial and symmetrical of their steering,” Nancy Fulda, a professor at Brigham Younger College, mentioned in a press release. “The outcomes present vital and repeatable constructive and unfavorable biases towards sure perception techniques.”