PolicyBank: Evolving Policy Understanding for LLM Agents

2026-04-16

AI Summary

This work addresses the challenge that large language model (LLM) agents often exhibit “compliant yet incorrect” behaviors when executing natural language organizational policies, due to policy ambiguity or logical flaws. To mitigate this, the paper introduces PolicyBank—the first structured policy memory mechanism supporting dynamic evolution. PolicyBank continuously refines policy understanding during pre-deployment testing by integrating tool-level policy insight extraction, interactive feedback learning, and iteratively updatable memory storage. This approach departs from the conventional paradigm of treating policies as static ground truths. Evaluated on a testbed with controllable vulnerabilities, PolicyBank significantly outperforms existing memory-based methods, closing up to 82% of the performance gap and achieving near-human expert levels in policy adherence and correctness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Mechanism Design
• Natural Language Processing: Safety and Robustness

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

LLM agents operating under organizational policies must comply with authorization constraints typically specified in natural language. In practice, such specifications inevitably contain ambiguities and logical or semantic gaps that cause the agent's behavior to systematically diverge from the true requirements. We ask: by letting an agent evolve its policy understanding through interaction and corrective feedback from pre-deployment testing, can it autonomously refine its interpretation to close specification gaps? We propose PolicyBank, a memory mechanism that maintains structured, tool-level policy insights and iteratively refines them -- unlike existing memory mechanisms that treat the policy as immutable ground truth, reinforcing "compliant but wrong" behaviors. We also contribute a systematic testbed by extending a popular tool-calling benchmark with controlled policy gaps that isolate alignment failures from execution failures. While existing memory mechanisms achieve near-zero success on policy-gap scenarios, PolicyBank closes up to 82% of the gap toward a human oracle.

Problem

• policy compliance
• specification gaps
• LLM agents
• natural language policies
• authorization constraints

Innovation

• PolicyBank
• LLM agents
• policy understanding
• memory mechanism
• specification gaps