The rapidly iterating Big Language Model in the disruptive transformation of AI2.0 has profoundly changed the research paradigm of natural language processing. Cross-domain model products with strong emergence capabilities are embedded in the main links of knowledge collection, mining, storage management, and shared applications of smart libraries. They can accurately and quickly deeply mine readers' behavioral patterns, demand characteristics, and emotional links in multi intention and multi-step of complex tasks, vigorously promoting content optimization, functional expansion, and mode innovation of library knowledge services. They also expose full-stack technical risks, ethical risks, regulatory risks, privacy risks, safety risks, copyright protection risks in the process of generic application. On the basis of building a security driven embedded model and a value aligned operational paradigm, it is urgent to enhance the technical defense and quality evaluation capabilities of model applications, improve relevant policies, regulations, regulatory measures, and cultivate professional skills of smart librarians, in order to promote rational, safe, reliable, and explicable innovation development of knowledge collaboration.