Large Language Models for Automatic Detection of Sensitive Topics