[Purpose/Significance] As artificial intelligence (AI) becomes increasingly embedded in organizational workflows, knowledge transfer is shifting from traditional human-to-human interaction toward a hybrid collaborative process of "human-AI-human." Conventional human-centric knowledge transfer models are often constrained by factors such as cognitive load, information overload, insufficient sharing motivation, and tacit knowledge loss, making it difficult to adequately support the preservation of organizational knowledge assets, the continuity of organizational memory, and process innovation. Therefore, this study constructs a theoretical framework for AI-mediated knowledge transfer from the perspective of knowledge management, explaining how AI, as a proactive sociotechnical intermediary, restructures the organizational processes of knowledge capture, sharing, comprehension, and reuse. [Method/Process] Employing a literature review and theoretical integration approach, and grounded in sociotechnical systems theory, the SECI model of knowledge creation, and the ability-motivation-opportunity (AMO) framework, this study systematically examined the process pathways, role reconfiguration, and operational boundaries of AI intervention in knowledge transfer. [Results/Conclusion] The findings indicate that AI primarily enhances knowledge accessibility, comprehensibility, and reusability through four key processes: knowledge capture and codification, matching and recommendation, translation and adaptation, and integration and reuse. Concurrently, it drives knowledge workers to assume two complementary functional roles: AI orchestration and knowledge stewardship. Building upon this, the study further proposes a dual-role reconfiguration model and, from a knowledge management perspective, reveals the sociotechnical restructuring logic underlying AI intervention in knowledge transfer, thereby offering theoretical guidance for organizations to construct human-AI collaborative knowledge management systems, preserve organizational memory, and foster continuous process innovation.