Various real-life planning problems require making upfront decisions before all parameters of the problem have been disclosed. An important special case of such problem especially arises in scheduling and staff rostering problems, where a set of tasks needs to be assigned to an available set of resources (personnel or machines), in a way that each task is assigned to one resource, while no task is allowed to share a resource with another task. In its nominal form, the resulting computational problem reduces to the well-known assignment problem that can be modeled as matching problems on bipartite graphs. In recent work \cite{adjiashvili_bindewald_michaels_icalp2016}, a new robust model for the assignment problem was introduced that can deal with situations in which certain resources, i.e.\ nodes or edges of the underlying bipartite graph, are vulnerable and may become unavailable after a solution has been chosen. In the original version from \cite{adjiashvili_bindewald_michaels_icalp2016} the resources subject to uncertainty are the edges of the underlying bipartite graph. In this follow-up work, we complement our previous study by considering nodes as being vulnerable, instead of edges. The goal is now to choose a minimum-cost collection of nodes such that, if any vulnerable node becomes unavailable, the remaining part of the solution still contains sufficient nodes to perform all tasks. From a practical point of view, such type of unavailability is interesting as it is typically caused e.g.\ by an employee's sickness, or machine failure. We present algorithms and hardness of approximation results for several variants of the problem.