AgentAbstain: Do LLM Agents Know When Not to Act?

2026-07-10

AI Summary

Current large language model (LLM) agents lack the ability to proactively abstain from acting when faced with ambiguous instructions, conflicting constraints, or tool failures, often leading to irreversible errors. This work introduces AgentAbstain, the first systematic evaluation framework for agent abstention, encompassing eight abstention scenarios and 263 paired tasks. It also presents AbstainGen, an automated pipeline for scalable, contamination-resistant synthesis of paired tasks through controlled perturbations of instructions, tools, and environment states. Task quality and reproducibility are ensured via deterministic replay, semantic evaluators, and human verification. Experiments across 17 prominent LLMs and four agent frameworks reveal that even the best-performing model (Gemini 3.1 Pro) achieves only 59.5% pairwise accuracy, with widespread failure modes such as post-hoc abstention. These findings demonstrate a critical deficiency in current agents’ abstention capabilities, which appear largely decoupled from general task-solving proficiency.

Abstract

Agent systems based on large language models (LLMs) are increasingly deployed for autonomous tasks, yet existing evaluations mostly focus on task success rather than whether agents know when to abstain. This gap poses real risks: under ambiguity, conflicting constraints, or tool failures, agents may execute unintended and irreversible actions. To close this gap, we present the first systematic evaluation framework for agentic abstention: the calibrated ability of tool-using LLM agents to recognize when not to act. At its core, AgentAbstain is a paired-task benchmark built on an agent-native taxonomy of 8 abstention scenarios across pre-execution reasoning and runtime discovery. It contains 263 paired tasks across 42 executable sandbox environments, where each pair consists of a should-act task and a should-abstain variant produced through a controlled perturbation 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, and three independent annotators rate 94-98% of sampled tasks as well-designed. Across 17 frontier LLMs in 4 agent harnesses, 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, in which agents execute irreversible actions before recognizing abstention triggers. Our code and dataset are open-sourced at agentabstain.github.io.

Problem

• LLM agents
• abstention
• autonomous tasks
• irreversible actions
• evaluation gap

Innovation

• agentic abstention
• paired-task benchmark
• AbstainGen
• tool-using LLM agents
• calibrated refusal