Large language models exhibit biases in answering context-dependent health questions, favoring specific demographic groups.
