Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset. AI and Wisdom, v. 1, n. 2, 1 Jun.2026.