According to Stanford AI Lab, research by @EchoShao8899 and @Diyi_Yang highlights a privacy concern where Language Models (LMs) leak sensitive information in over 30% of cases when performing tasks, despite understanding privacy norms in question-answering scenarios.