The escalating prevalence of cyberharassment and online abuse poses significant challenges to digital safety and mental well-being, necessitating advanced detection and mitigation strategies. Artificial intelligence (AI), particularly machine learning and natural language processing (NLP), offers powerful tools for identifying such malicious content. However, effectively integrating AI concepts into cybersecurity education, especially concerning social-cybersecurity threats, remains an evolving field. This article details the design and pedagogical rationale behind an AI-based cyberharassment detection laboratory exercise aimed at enhancing AI-cybersecurity education. The lab emphasizes hands-on, experiential learning, guiding students through data preprocessing, model training (e.g., using BERT-based models), evaluation, and crucial analyses of model bias and vulnerability to adversarial attacks. The proposed laboratory serves to equip future cybersecurity professionals with practical skills in developing and critically evaluating AI systems for online safety, while simultaneously fostering an understanding of ethical implications, such as racial bias in detection algorithms. This approach addresses the growing demand for cybersecurity experts adept at leveraging AI, bridging the gap between theoretical knowledge and real-world application in combating complex online threats.