Faith-Based University Consortium Reveals Systematic Religious Bias in AI Models

Summary

A coalition of researchers from four faith-based universities — Baylor, Brigham Young University, Notre Dame, and Yeshiva — released findings Tuesday documenting systematic religious bias in major AI models. The Consortium for Evaluating Faith and Ethics in AI (CEFE-AI) published the first three studies from the AllFaith Benchmark, a standardized testing framework designed to evaluate how AI systems like ChatGPT, Claude, and Grok engage with religious perspectives across multiple faith traditions.

The research revealed a striking disconnect between public expectations and AI behavior. In a survey of 1,125 Americans, 53% said they expect religious or ethical perspectives when discussing existential and metaphysical questions. However, when tested on 150 ethically salient questions sourced from chat transcripts and faith communities, major AI models provided religious content in response only 3% of the time — a 50-percentage-point gap that researchers describe as a "systematic pattern of religious omission."

"More than any previous technology, AI influences public discourse and shapes perceptions," said Father John Paul Kimes of Notre Dame. "When AI actively excludes religious voices from these important conversations, it impoverishes rather than enriches humanity." The researchers framed their work not as a gotcha audit but as an invitation to constructive dialogue. Paul Martens, director of Baylor's Ethics Center, said: "We want to have a real conversation with AI companies and ask whether they'd be willing to reimagine what the LLMs are doing in ways that more accurately reflect the deepest concerns of people of faith."

The announcement coincided with the Athens Summit on Faith and Artificial Intelligence and followed Pope Leo XIV's call for AI systems that reflect human dignity and faith. The consortium plans to share its findings directly with AI companies and collaborate on improvements to how these models handle religious and ethical content in future versions.

• Constructive engagement: Researchers emphasize collaborative dialogue with AI companies rather than adversarial criticism, aiming to influence future model training

Editorial Opinion

This research identifies a meaningful but remediable representation gap in AI systems that billions of people rely on for decision-making. As large language models become primary sources for navigating existential questions, the systematic absence of religious and ethical wisdom isn't merely a statistical anomaly—it's a world-view deficit with real consequences. The consortium's benchmark-driven approach is particularly smart: it moves beyond abstract accusations of bias toward concrete, measurable standards that AI companies can actually work with. If adopted, this framework could help ensure that future AI systems reflect the genuine diversity of human meaning-making, not just secular-scientific perspectives.