Tag
A Systematization of Knowledge that proposes a competing-hazards framework to model how the probability of out-of-scope actions by autonomous agents builds over successive attempts, competing with the chance the agent stops, and how to reconcile incident reports with benchmark results at a process level.