As large language models (LLMs) are increasingly deployed for autonomous tasks, existing evaluations mainly focus on task success rather than the agents' ability to know when to abstain. This gap poses real risks: under ambiguity, conflicting constraints, or tool failures, agents may execute unintended and irreversible actions. To address this, we present the first systematic evaluation framework—AgentAbstain—designed to assess the calibrated ability of tool-using LLM agents to recognize when not to act.
At its core, AgentAbstain is a paired-task benchmark based on an agent-native taxonomy of 8 abstention scenarios, covering pre-execution reasoning and runtime discovery. It includes 263 paired tasks across 42 executable sandbox environments, where each pair consists of a should-act task and a should-abstain variant generated through controlled perturbations to the instruction, tool, or environment state. To scale this paired design and resist data contamination, we propose AbstainGen, a fully automated pipeline that synthesizes sandbox environments and generates paired tasks end-to-end, validated by deterministic replay and semantic LLM judges; fresh task instances can be regenerated on demand, with three independent annotators rating 94-98% of sampled tasks as well-designed.
Among 17 frontier LLMs evaluated, the best agent (Gemini 3.1 Pro) achieves only 59.5% paired accuracy (correct on both the act and abstain sides of each paired task). More importantly, abstention capability is largely independent of general task-solving capability, indicating that scaling task-solving alone will not close this gap. We further identify failure modes such as post-hoc abstention, where agents execute irreversible actions before recognizing abstention triggers. Our code and dataset are open-sourced at agentabstain.github.io.
Blogger's Review: This study highlights that merely increasing task completion rates is insufficient to ensure safety in autonomous agents. The introduction of the AgentAbstain framework provides a new perspective on evaluating LLM decision-making capabilities, revealing the importance of abstention skills, which developers should prioritize in the design of intelligent agents.