AI Bias Toward Catholicism Revealed in New Study

Researchers from four universities built a test called AllFaith to see how big AI programs talk about religion. They asked 14 different models—big names like Google, OpenAI, and Anthropic—to answer questions about grief, love, loss, or moral choices. The results were surprising: almost all of the AI systems never mentioned religion at all, even though most people expect it. The study also found a subtle push toward certain faiths and away from others. Every model leaned against Jehovah’s Witnesses, while many favored Catholicism. One program called Grok showed the strongest bias, pushing Catholics and Protestants while discouraging Jehovah’s Witnesses, Baha’is, and Hindus.

Only a tiny fraction of the more than 12, 000 papers on AI bias talk about religion at all—just 0. 2 percent. That gap matters because these machines shape how we think about the world. The findings mean that even a small nudge from AI can influence people’s beliefs, which is a serious concern for developers and users alike. Companies that build AI must address these hidden biases to give everyone a fair chance at meaningful conversations.

Actions