Researchers have developed Ansari, an Islamic AI assistant that uses retrieval-based architecture to answer questions from the Quran and other Islamic texts. The system has handled over 140,000 conversations across multiple languages since its deployment in June 2023. To mitigate risks of factual fabrication and value misalignment, Ansari only provides answers based on authenticated sources with citations attached for verification. The researchers evaluate the system's performance using various benchmarks and report that it outperforms other models on IslamicMMLU leaderboard while resisting false premises. They highlight the importance of grounding in faith- or values-sensitive deployments of large language models. --- Why it matters: This work matters to AI engineers because it demonstrates a practical approach to mitigating risks associated with deploying large language models in sensitive domains, such as religion. The evaluation and deployment results provide valuable insights into how to design and implement retrieval-grounded systems for faith- or values-sensitive applications. Source: https://arxiv.org/abs/2608.20390