Faith researchers build AI benchmarks to test religious bias

The G20 Interfaith Forum will host a July 2 webinar on how computer scientists and faith scholars are measuring whether AI systems misstate, stereotype or omit religion. The effort includes a new multi-university consortium aimed at creating benchmark tests for faith, ethics and human-flourishing questions.

Why it matters: - AI tools are increasingly shaping how people learn about faith, making accuracy and fairness in religious content a practical issue, not just a theological one. - Researchers say better benchmarks can reveal religious bias, stereotypes, secularization and missing references to faith in large language models. - The goal is to give developers, universities and governments a way to measure whether AI supports human flourishing in religious contexts.

What happened: - The G20 Interfaith Forum will host “AI and Benchmarks” on Thursday, July 2, 2026, at 12:00 PM EDT. - The webinar is the third event in the forum’s eight-part AI and Human Flourishing series. - The session will be held virtually and is free to attend. - Registration is available through the webinar link. - The panel will bring together computer scientists and faith-based researchers to discuss how benchmarks can evaluate AI behavior around religion.

The details: - David Wingate said AI benchmarks can quantify ways AI may be enhancing or diminishing human flourishing. - Wingate said benchmark designers can measure religious bias, negative stereotypes, multilingual consistency and omission of religion where it would be expected. - Wingate argued governments can support benchmark design and maintenance to generate actionable insights into AI behavior. - Wingate said academically rigorous benchmarks can help work with AI providers to ensure AI respects religion and its role in society. - The Consortium for the Evaluation of Faith and Ethics in AI, or CEFEAI, is building benchmarks focused on honest, accurate and respectful representations of faith, religion and ethics. - CEFEAI is a pluralistic, multi-university consortium of faith-based research institutions. - Initial consortium members include Baylor University, Brigham Young University as lead, the University of Notre Dame and Yeshiva University. - The consortium plans additional representation across more religious communities. - Participating universities are building multiple benchmarks, including faith-specific expertise, question design, evaluation rubrics and shared infrastructure. - The effort is intended to measure bias, negative stereotypes, secularization and other emergent phenomena in AI. - CEFEAI aims to create an ongoing, multi-faith benchmark that can serve as a gold-standard assessment for large language model safety and truthfulness in religious contexts. - The webinar speakers include David Wingate, Michael Graham and Marianna Richardson. - Wingate is a professor of computer science at Brigham Young University and works at the intersection of machine learning and social science. - Graham is program director of the Keller Center for Cultural Apologetics at The Gospel Coalition, founder of the AI Christian Benchmark and co-founder of Religion Data. - Richardson is director of communications for the G20 Interfaith Forum and an adjunct professor of management communication at the BYU Marriott School of Business. - The G20 Interfaith Forum said IAMC is a co-sponsor of the event. - The forum says it works with religious thought leaders and political representatives to help shape the G20 agenda. - The forum also says it draws on religious institutions and beliefs to help prioritize global policy goals and practical implementation. - More information is available at g20interfaith.org.

Between the lines: - The push for religion-focused benchmarks reflects a broader shift in AI evaluation from narrow technical accuracy to cultural and social accountability. - The emphasis on pluralistic, community-grounded testing suggests faith communities want a seat at the table before AI norms harden. - The consortium’s focus on measurement, not theological resolution, signals an attempt to keep the work empirical rather than doctrinal.

What's next: - CEFEAI and its university partners are expected to keep building and standardizing benchmarks across religious traditions. - The July 2 webinar will likely serve as a public milestone for the project and a forum for future collaboration. - Researchers say the long-term aim is an ongoing benchmark that can track how well AI systems handle religious content over time.

The bottom line: - Faith groups and AI researchers are turning religious representation in AI into something measurable, hoping benchmarks will pressure model makers to reduce bias and improve truthfulness.