> Researchers found that every AI model tested showed positive bias toward Catholicism and negative bias toward Jehovah’s Witnesses.
Did they try using other languages? This sounds like a correlation with publicly-available English-language text in specific.