Insufficient examination of safety risk propagation within complex construction projects (CCPs) leads to defective risk responses. This study attempts to exemplify metro deep foundation pit projects to investigate the mechanisms underlying this propagation. The interaction network of safety risk factors is constructed utilizing association rule mining and link prediction. A modified susceptible-infectious-recovered (SIR) model is used to simulate and analyze risk propagation within the network. The results highlight distinct risk propagation dynamics triggered by various initial hazardous/controlled factors. Risk factors from the owner or social environment are critical in initiating and controlling propagation, while those from the survey unit, designer, supervision unit, monitoring unit, working environment, or the contractor's organizational and managerial aspects have moderate effects. The contractor-related unsafe behaviors, mechanical, material, or structural factors, and natural environmental factors exhibit relatively limited impact. The ultimate risk level is contingent upon initial hazardous and controlled factors, with the risk controllability solely determined by the latter. It is recommended that the focus of safety management be broadened from contractors to owners and the social environment, and from isolated risk responses to risk network management. This study may deepen managers' understanding of the sources of safety risk within CCPs, thereby promoting project success.