A research paper introduces ObligationBench, a benchmark for evaluating the capability of guard models to identify required yet unfulfilled safety-critical actions (obligations) in LLM-based agents, and ObligationGuard, a model that surpasses existing guard models in identifying these obligations.