POLAR: Ontology-Guided Risk Prevention for Tool-Calling LLM Agents

2026-10-06

AI Summary

This study addresses the absence of preemptive risk prevention mechanisms in LLM tool agents by proposing POLAR, a novel defense framework. POLAR introduces a dual-layer ontology-based reversibility grading score and a candidate inverse sequence derivation mechanism. By leveraging structured ontological modeling to assess operation reversibility, the framework prunes high-risk tool calls prior to execution and supports integration with small-model agents. Experimental evaluations on τ²-bench demonstrate that POLAR effectively improves task rewards for specific scenarios while delineating the trade-off boundary between utility and safety. Ultimately, this work provides auditable, preemptive safety guardrails for LLM agents, mitigating risks before irreversible actions occur.

Abstract

LLM tool-use agents operate in dynamic environments where many actions carry operational risk. However, most safety mechanisms react only after errors manifest. Existing pre-emptive approaches either fine-tune the agent on chain-of-thought deliberation or compile natural-language guardrails into runtime checks, but they do so without exposing a structural, auditable verdict. We propose POLAR, a guardrail framework for small tool-calling agents that assesses reversibility through a structured two-layer ontology. POLAR assigns each action a graded reversibility score by deriving a candidate inverse sequence; calls failing a threshold are pruned before execution. Evaluated on $τ^2$-bench across six agent models, POLAR improves mean task reward by 0.11 to 0.18 points on airline for four of six agents, but only eight of eighteen model--domain cells improve overall; retail and stronger agents often regress. POLAR provides an auditable structural check and characterizes its task-utility trade-offs. Reward is not a direct measure of prevented harm.

Problem

• LLM agents
• tool-calling
• risk prevention
• safety guardrails
• action reversibility

Innovation

• Ontology-Guided Guardrail
• Action Reversibility
• Tool-Calling Agents
• Preemptive Risk Prevention
• Auditable Safety