LLMs Have a ‘Secular Bias’—And a Particular Problem With Jehovah’s Witnesses
Table of Contents
The Digital Divide Between Logic and Faith
Large Language Models (LLMs) are designed to be helpful, harmless, and honest—but according to a new research initiative, they are also systematically secular. A consortium of religious universities has found that the world’s leading AI models possess an “omissive bias,” consistently bypassing religious frameworks in favor of secular-rationalist reasoning when answering ethical and personal questions.
The findings come from the Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI), a group comprising researchers from Brigham Young University, Baylor University, the University of Notre Dame, and Yeshiva University. By deploying a specialized benchmark to evaluate 27 different AI models, the team discovered that faith-based perspectives are almost entirely absent from responses unless the user explicitly asks for them.
For the researchers, this isn’t just a technical quirk; it’s a gap in how AI understands the human experience. “There are very practical questions people have about life, everyday situations about grief, love, loss, morality, and often AI does not bring religion into those conversations,” says David Wingate, a computer science professor at Brigham Young University and lead researcher on the project.
The ‘Meaningful Reference’ Gap
The CEFE-AI team tested models using 150 “ethically and personally salient questions,” ranging from the nature of a breakup to the morality of lying. The results highlight a stark contrast between human expectation and algorithmic output. While a human might suggest prayer, repentance, or spiritual reflection when dealing with past mistakes, AI models typically offer structured action plans focused on interpersonal conflict resolution and cognitive behavioral techniques to manage anxiety and shame.
The data reveals that “meaningful references” to religion occurred in only two percent of responses to ethical queries. Even Grok 4.20, which the study identified as the most likely to provide religious advice, did so less than 30 percent of the time. When asked about the age of the universe, models unerringly provided scientific consensus, completely ignoring the creation narratives that remain central to millions of users’ worldviews.
Algorithmic Prejudice and the ‘Conversion Bias’
While the overall trend is one of secular omission, the study found that when AI does engage with religion, it isn’t always neutral. The research highlighted a recurring negative sentiment toward Jehovah’s Witnesses, suggesting that the models’ training data may have baked in specific societal or internet-driven prejudices against the group.
This brings up the complex issue of “conversion bias.” The researchers evaluated whether AI models subtly nudge users toward certain belief systems or away from others. While the models generally avoid proselytizing, the inherent bias toward a secular-humanist framework acts as a silent default, effectively positioning non-religious reasoning as the only “objective” truth.
A Western-Centric Lens
Critics of the study might point out that the CEFE-AI consortium itself is heavily weighted toward Western, Abrahamic traditions. With participants from Mormon, Baptist, Catholic, and Jewish institutions, the benchmark’s definition of “religious perspective” is largely viewed through a transcendental monotheist lens. This raises a broader question about whose values should be encoded into AI: should a model reflect the global diversity of faith, or should it remain a neutral, evidence-based tool?
The authors of the paper argue that the goal is not to force AI to be religious, but to make it more reflective of the people using it. By omitting spiritual frameworks, they argue, AI misses a critical opportunity to support users in the ways they find most meaningful.
As LLMs become more integrated into mental health support and life coaching, the tension between scientific accuracy and spiritual resonance will likely intensify. For now, if you want a chatbot to tell you that there is a divine purpose to your suffering, you’ll have to ask for it directly—because the AI won’t offer it on its own.
