Abstract: This study develops and empirically evaluates an artificial intelligence (AI) literacy framework for social workers in response to the transformation of professional competencies in social work under the conditions of algorithmic governance. It aims to provide theoretical insights and practical tools for addressing the disconnect between technological development and social work practice and for improving the precision of public service delivery. Drawing on Bloom's Cognitive Taxonomy, the study first develops a four-dimensional conceptual framework encompassing AI awareness, AI skills, AI application, and AI ethics. The Delphi method is then employed to construct an AI literacy assessment framework for social workers, followed by an empirical investigation and statistical analysis based on data collected from 303 social workers. The findings indicate that social workers in China demonstrate a moderately above-average level of AI literacy, with significant positive interrelationships among different dimensions. No significant differences were found across demographic groups, suggesting that AI literacy constitutes a broadly applicable professional competency. However, the study also identifies a structural imbalance characterized by relatively strong AI application capabilities but insufficient ethical awareness, indicating a potential risk of instrumental rationality overriding value rationality. Based on these findings, the study proposes four pathways for enhancing AI literacy among social workers: prioritizing ethical integration, embedding AI applications in professional practice contexts, implementing tiered and progressive training programs, and fostering an inclusive AI-enabled ecosystem across the social work profession. These strategies aim to strengthen the integration of ethical principles, professional practice, and technological capabilities, thereby promoting the sustainable development of AI literacy in social work.