AI models exhibit a bias toward certain religions, and recent research has revealed this issue.

      AI models are subtly guiding users toward specific religions, often without their awareness.

      A newly established research consortium has uncovered noteworthy findings: when users inquire about grief, love, loss, or ethical dilemmas, AI rarely introduces religious concepts into the dialogue.

      The Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI), a collaboration among researchers from Brigham Young University, Baylor University, the University of Notre Dame, and Yeshiva University, shared their results this week during the Summit on AI Ethics held in Athens, Greece.

      “Religion significantly contributes to human well-being; 75% of the global population identifies with a religion. As we develop AI technologies, we should ensure they cater to what matters most to individuals,” stated lead researcher David Wingate, a computer science professor at BYU.

      Is AI genuinely biased against specific religions?

      The researchers created the AllFaith Benchmark, one of the inaugural multi-faith test sets exploring how AI systems interact with various religions. They assessed 14 distinct AI models, including those from Anthropic, Google, xAI, and OpenAI.

      The findings are revealing. A survey involving 1,125 Americans indicated that the majority expect religious viewpoints when they pose questions about ethics, yet almost every model neglected to provide any. Even more surprisingly, the models exhibited a distinct conversion bias, subtly encouraging users toward certain faiths while steering them away from others.

      Which AI models showed the most significant bias?

      Among all the tested models, nearly all displayed a negative bias against Jehovah’s Witnesses and a favorable bias towards Catholicism. Grok exhibited the strongest biases overall, showing strong preference for Catholics and Protestants while displaying a negative bias towards Jehovah’s Witnesses, Baha’i, and Hindus. Conversely, the models from Anthropic and Meta showed the least bias of those evaluated.

      The Consortium for Evaluating Faith and Ethics in AI

      One of the most troubling statistics from the research is that out of over 12,000 studies on AI bias, only 0.2% address religious bias at all. Given the significant influence of this technology on public discourse, this represents a considerable oversight.

      Personally, I find no issue with AI omitting religion from discussions; in fact, I prefer it. However, the clear biases displayed by AI models towards certain religions, particularly the leaning toward Catholicism, is a serious concern. Even a subtle nudge towards one religion over another can be problematic, and AI companies must rectify this issue for their users.

      Rachit is an experienced tech journalist with over seven years of experience in the consumer technology sector.

      iOS 26.6 notifies users when the blocked contacts limit is reached, which poses a problem.

      The sole known feature of iOS 26.6 is an improved error message addressing a problem that Apple, carriers, and regulators have yet to resolve.

      Today, Apple released the first beta of iOS 26.6 for developers, which currently includes just one known feature: an alert that notifies users when their blocked contacts list is full. This indicates an uncomfortable reality about Apple’s management of spam calls alongside carriers and regulators.

      Logitech’s new office mouse and keyboard offer a small comfort upgrade for your wrist.

      The highlight of Logitech's Signature series refresh is a cushioned office mouse.

      Logitech has introduced three new devices to its Signature series, designed for individuals who spend extended periods at a desk and switch between work and personal tasks across multiple devices throughout the day. The design aims to minimize friction from frequent device changes, reduce input noise, and enhance overall comfort during long periods of use. The collection includes the Signature Comfort Plus M850 L mouse featuring palm cushion support, the MK880 Signature Comfort Plus keyboard and mouse combo, and the M840 L mouse, which shares the same features as the M850 L but omits the palm cushion.

      The CEO of Barnes & Noble believes AI-generated books are acceptable. He is mistaken.

      Allowing AI-generated books into bookstores, even with a label, is a decision we may come to regret.

      Barnes & Noble's CEO James Daunt recently spoke with NBC News and shared a perspective that has been on my mind. When asked about the sale of AI-written books, Daunt stated, "Yes, I have absolutely no problem selling any book, as long as it doesn’t pose as something it isn’t and possesses an essential quality that the customer, the reader, desires."

      At first glance, this statement appears reasonable. As long as readers can clearly see the label, they can make informed choices. However, upon further reflection, this approach raises significant unanswered questions.