In the research field of grid risk management, a combination of information systems and manual work is usually used. As the power grid continues to evolve, the use of artificial intelligence (AI) is gradually increasing, the patterns of efficient Human-AI collaboration become a crucial research problem. Management interface risks in the power grid industry are the type of risks that are likely to occur during the project feasibility study review stage. Once these risks occur, they will have a certain degree of impact on each business process node in the subsequent stages of the project. This study aims to explore how these risks can be managed through effective collaboration between human and AI. First, this paper examines human-computer division of labor criteria for management interface risks, with a particular focus on rule-driven tasks. Second, we propose a “composite algorithm”, which integrates cosine similarity, BERT similarity, hybrid similarity and APRIORI algorithm. This algorithm enables risk rule extraction, risk keyword extraction, risk structure matching and risk correlation identification, ultimately constructing a risk rule base. Finally, the effectiveness of this human-AI collaborative risk management method is verified using an enterprise in the power sector as a case study. This paper is innovative at the field level, the application level and the organizational management level.