Can AI be truly objective? A fascinating new study reveals how large language models (LLMs) develop cultural biases, raising questions about their fairness and trustworthiness. Researchers tested four popular LLMs—GPT-3.5-Turbo, GPT-4o-mini, Llama 3.1, and MistralNeMo—across eight languages, using the Moral Foundations Questionnaire (MFQ-2) to gauge their moral reasoning. The MFQ-2 assesses moral values like care, fairness, loyalty, and authority, allowing comparisons between AI and human moral judgments across cultures. The study found substantial variations in how different AI models prioritized these values depending on the language used, challenging the assumption of universal moral consistency in AI. Interestingly, the research disproved the theory that LLMs impose English-centric moral norms due to training data dominance. While English influenced moral judgments, the models adapted surprisingly well to different cultural contexts, demonstrating the complexity of AI's moral development. However, significant differences emerged between Western and non-Western language groups, suggesting cultural biases within the models themselves. Notably, larger, more data-rich models like GPT-3.5 and GPT-4o-mini aligned more closely with human responses than smaller models, particularly in well-represented languages like English. This suggests data diversity is crucial for creating culturally sensitive AI. The study's findings have major implications for building trustworthy AI. It highlights the urgent need for culturally inclusive development practices to ensure fairness and avoid unintentional biases. As AI's role in society expands, addressing these cultural nuances is vital for building a future where AI benefits everyone, regardless of their background.