Agentic Abstention: Do Agents Know When to Stop Instead of Act?
Abstract
We study agentic abstention, the ability of tool-using LLM agents to recognize when a task is infeasible, contradictory, or missing unrecoverable information and stop instead of continuing to act.
Type
Publication
NeurIPS 2026 Submission
We study agentic abstention: when tool-using LLM agents should recognize that a task is infeasible, contradictory, or missing information that cannot be recovered, and stop instead of spending additional turns on futile actions.
