AI models systematically fail at Christian moral reasoning — 31-point drop in spiritual questions

Researchers tested 20 major AI models against Christian values and found they perform worse on spirituality and moral questions than on other tasks, dropping 17 points on average across all seven dimensions measured. This suggests AI companies train their models to be inoffensive to everyone, which means they're incoherent to anyone with a specific worldview.

For years, AI companies claimed their models were neutral tools — just reflecting the average of their training data. This paper shows that's false. The systems are optimized for procedural blandness, which works fine for math problems but collapses when you ask them to reason about meaning, purpose, or spiritual life. If you care about theology, philosophy, or any coherent moral framework, the AI you're using was trained to not care about the things you do.