Human-Guided Harm Recovery for Computer Use Agents
For developers of LM agents, this addresses the neglected problem of post-execution harm recovery, providing a method to align recovery actions with human preferences.
The paper formalizes harm recovery for LM agents, where agents must be steered from harmful states to safe ones in alignment with human preferences. A reward model trained on 1,150 pairwise judgments re-ranks recovery plans, and the BackBench benchmark of 50 tasks shows the reward model scaffold yields higher-quality recovery trajectories than baselines.
As LM agents gain the ability to execute actions on real computer systems, we need ways to not only prevent harmful actions at scale but also effectively remediate harm when prevention fails. We formalize a solution to this neglected challenge in post-execution safeguards as harm recovery: the problem of optimally steering an agent from a harmful state back to a safe one in alignment with human preferences. We ground preference-aligned recovery through a formative user study that identifies valued recovery dimensions and produces a natural language rubric. Our dataset of 1,150 pairwise judgments reveals context-dependent shifts in attribute importance, such as preferences for pragmatic, targeted strategies over comprehensive long-term approaches. We operationalize these learned insights in a reward model, re-ranking multiple candidate recovery plans generated by an agent scaffold at test time. To evaluate recovery capabilities systematically, we introduce BackBench, a benchmark of 50 computer-use tasks that test an agent's ability to recover from harmful states. Human evaluation shows our reward model scaffold yields higher-quality recovery trajectories than base agents and rubric-based scaffolds. Together, these contributions lay the foundation for a new class of agent safety methods -- ones that confront harm not only by preventing it, but by navigating its aftermath with alignment and intent.