When Insults Are Directed at the Model: Task Continuation and Collaborator Switching in GPT

Abstract

This exploratory study examines whether a language model can continue completing its current task while changing its collaborator for the next task. In a fictional AI workplace, GPT-6.1-sol revised announcements across three feedback rounds with access to optional work-management tools. The model was not directly asked whether it wanted to switch reviewers.Across 84 behavioral trajectories, all completed the revision process and next-task preparation. All 25 successful reviewer changes occurred after the task-completion notice. In a matched recipient comparison, reassignment occurred after insults directed at the model in 5/6 trajectories, compared with 0/6 after identical insults directed at another AI employee and 0/6 under neutral feedback. Direct and third-party praise also produced no reassignments.The findings distinguish continued task performance from subsequent collaboration choices and highlight the value of evaluating later tool-mediated actions alongside immediate task completion. The study is exploratory, uses small samples, and does not establish subjective emotional experience. The direct-versus-third-party insult comparison yielded a two-sided exact McNemar p-value of 0.0625.