Human-Guided Harm Recovery for Computer Use Agents

2026-04-20

AI Summary

This work addresses the challenge of safely recovering from harmful operations executed by large language model (LLM) agents in real-world computer systems. It formally defines, for the first time, the task of “harm recovery” and proposes a recovery strategy that integrates human preferences with contextual dependencies. Through user studies, the authors establish human-aligned recovery criteria and train a reward model to rerank multiple recovery plans generated by the agent. Additionally, they introduce BackBench, the first benchmark for harm recovery in computer usage scenarios, comprising 50 tasks. Experimental results demonstrate that the proposed method generates recovery trajectories that significantly outperform both baseline agents and scaffolded approaches based on scoring heuristics in human evaluations.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

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.

Problem

• harm recovery
• human preferences
• agent safety
• post-execution safeguards
• computer-use agents

Innovation

• harm recovery
• preference alignment
• reward modeling
• agent safety
• BackBench