Drug toxicity prediction is a cornerstone of modern drug design, focusing on assessing the harmful effects of chemical substances on biological systems. Traditional toxicology methods, reliant on in vivo and in vitro experiments, are costly and time-intensive and raise ethical concerns. Computational toxicology has emerged as a complementary and efficient approach, utilizing machine learning (ML) and deep learning (DL) techniques to predict toxicological profiles. These models enable early identification of toxicity risks, reducing reliance on animal testing and accelerating drug development. ML techniques like random forests and support vector machines use chemical descriptors for toxicity modeling but face challenges with limited datasets and overfitting. In contrast, DL methods such as graph neural networks (GNNs) and convolutional neural networks (CNNs) excel in handling large-scale data and automatically extracting intricate molecular features. These models have been applied to predict various toxicological endpoints, including hepatotoxicity, cardiotoxicity, and endocrine disruption. Furthermore, AI-driven approaches enhance mechanistic insights, enabling regulatory compliance and safety evaluation. As toxicity databases expand and algorithms evolve, AI-based toxicity prediction is poised to become integral to pharmaceutical innovation. Efforts to improve model interpretability, integrate biological pathways, and address dataset limitations are critical for advancing the reliability and applicability of these tools.