Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset. (2026). AI and Wisdom, 1(2). https://www.aiandwisdom.org/index.php/aiandwisdom/article/view/14