AI models have a religion favoritism problem, and new research exposes it

A university-led consortium tested 14 AI models with the AllFaith Benchmark on sensitive life topics.

The models often left religion out of moral or personal advice, even when faith would be relevant.

Researchers found uneven treatment across religions, with negative bias toward Jehovah’s Witnesses and positive bias toward Catholicism in many models.

The findings point to a major blind spot in AI safety and fairness work: religious bias is still underexamined.