Drug safety spans the full lifecycle of drug research and development, manufacture, distribution and usage, serving as both the cornerstone of public health maintenance and a core issue in regulatory science. Traditional drug safety monitoring models face numerous challenges regarding data processing efficiency, agility in risk identification and comprehensiveness. However, the advancement of artificial intelligence (AI) technologies offers unprecedented opportunities for this field. From a full lifecycle perspective, this paper systematically reviews and analyzes the key technologies and application progress of AI in safety assessment across three stages: preclinical research, clinical trials, and post-marketing pharmacovigilance. Currently, AI technologies, represented by machine learning and natural language processing, have demonstrated significant potential in drug toxicity prediction, the automated identification of adverse events, and real-time safety monitoring. Nevertheless, uneven data quality, insufficient model interpretability, and imperfect regulatory compliance remain the primary bottlenecks hindering the widespread application of AI in the field of drug safety. Future development will focus on the fusion of multi-modal data, the enhancement of model interpretability, and the application of federated learning for privacy protection, with the aim of constructing a more precise and reliable intelligent drug safety monitoring system.