Large language models exhibit linguistic discrimination, providing inconsistent responses and safety levels across different languages due to uneven training data distribution.
