IslamTrust Project Presented at MusIML Workshop Co-located With NeurIPS 2025

We are pleased to announce that our researchers presented their latest work, IslamTrust, at the 5th Muslims in ML (MusIML) Workshop co-located with the NeurIPS 2025 conference.

Authored by our thesis student Abderraouf Lahmar, and in collaboration with our colleagues Md. Easin Arafat and Zakarya Farou, and Mufti Mahmud (external), the project introduces a benchmark designed to evaluate the alignment of Large Language Models (LLMs) with Islamic values. The dataset consists of 406 multi-choice questions categorized across six domains, such as misconceptions, ethics, and everyday decision-making - in both Arabic and English. The study highlights that while LLMs perform better in Arabic, they still struggle with accurate alignment regarding Islamic ethics.

The paper can be found here.

More information on the workshop series can be found here.