Probabilistic Safety/Risk Assessment (PSA/PRA) models capture a vast amount of interconnected information. Basic Events typically represent equipment failures as well as human failure events. The failure behavior of components is characterized by mathematical reliability models with component-specific parameters. These events may belong to common cause failure groups, appear as leaf nodes in fault trees or define branching conditions in event trees. The structure of fault and event trees reflects how failures propagate through the plant, how systems and plant personnel respond and how undesired events are mitigated. This structure can also change depending on the analyzed scenario. When performed manually, development, maintenance, and customization of these models for various applications require thousands of hours of expert work.
In this paper, we discuss the automation of these activities, with particular emphasis on establishing trust in the quality of results obtained by automated methods. A range of tools can support this effort. Model-based techniques and Generative Artificial Intelligence (AI) can automatically construct fault trees from system engineering descriptions. Scripts, batch commands, and AI prompts enable large scale updates across the model. Connecting a PSA model directly to a plant information system might streamline probabilistic risk monitoring. However, these automated approaches also reduce direct user control over the final outcome. We outline pros and cons of different methods and propose a practical strategy for introducing automation into PSA workflows while maintaining confidence in model accuracy.