Large language models (LLMs) like ChatGPT have taken the world by storm, but can we really trust them, especially when we venture beyond English? A groundbreaking research paper introduces "XTRUST," a benchmark designed to put LLMs' multilingual trustworthiness to the test. Imagine asking an AI for medical advice in Hindi or legal guidance in Arabic. The reliability of the answer could have serious real-world consequences. XTRUST dives into this challenge by examining how LLMs perform across ten different languages, including Arabic, Chinese, French, German, Hindi, Italian, Korean, Portuguese, Russian, and Spanish. The researchers explored critical areas like identifying illegal activities, spotting AI "hallucinations" (where AI fabricates information), handling sensitive topics like mental and physical health, navigating biases, and resisting manipulation to spread misinformation or reveal private data. They tested popular LLMs including GPT-4, ChatGPT, and others. The results are a mixed bag. While GPT-4 generally performed well, some LLMs struggled significantly with lower-resource languages like Arabic and Russian. This highlights a key challenge: ensuring AI is equally trustworthy for all languages, not just those with vast amounts of training data. The findings also reveal a concerning gap. While LLMs are becoming incredibly sophisticated, the techniques for making them trustworthy haven't kept pace. As LLMs become more powerful, the risk of harmful or biased outputs increases, especially in multilingual settings. XTRUST marks an essential step in the quest for truly trustworthy AI. By pinpointing the weaknesses in current LLMs, it provides a roadmap for future development, paving the way for AI that's both powerful and reliable in every language.