Risk-Aware LLM Agents for Geospatial Data Retrieval: Design and Preliminary Adversarial Evaluation

2026-06-12

AI Summary

This study addresses the challenge of securely and accurately translating natural language queries into geospatial API calls by proposing a risk-aware multi-agent framework. The framework integrates three collaborative modules—Guardrail, General-QA, and Recommender-Analyst—to ensure semantic alignment while incorporating a prompt-layer interception mechanism that enhances both cross-platform portability and system-level defense capabilities. By synergistically combining large language models with a multi-agent architecture, the approach supports remote sensing data retrieval in critical applications such as environmental monitoring and disaster response. Empirical evaluation through adversarial multi-turn interactions demonstrates the effectiveness of the safety-oriented instructions, significantly improving system robustness and highlighting the urgent need for adaptive defense mechanisms in high-impact API manipulation scenarios.

Abstract

We present an LLM-driven framework for retrieving remote sensing data from cloud-based geospatial catalogues using natural language queries. The system converts user intent into structured API calls, enabling efficient access to satellite imagery and environmental datasets. The architecture integrates three agents: Guardrail for safety and policy enforcement, General-QA for intent interpretation, and Recommender-Analyst for schema-aware API call generation. This coordinated design ensures reliable, semantically aligned interaction with external data services. The modular framework is portable across platforms through API schema substitution and supports applications in environmental monitoring, disaster response, and climate analysis. It establishes a scalable interface between user intent and geospatial infrastructure, enabling streamlined and automated Earth observation workflows. Preliminary experiments under adversarial multi-turn settings show that prompt-level safety instructions improve robustness, although rare high-impact failures persist in API manipulation scenarios and highlight the need for adaptive, system-level defenses that balance safety, usability, and cost efficiency, which motivates the use of our intercept-level Guardrail agent.

Problem

• Risk-Aware
• LLM Agents
• Geospatial Data Retrieval
• Adversarial Robustness
• API Safety

Innovation

• Risk-Aware LLM Agents
• Geospatial Data Retrieval
• Modular Multi-Agent Architecture
• Adversarial Robustness
• Natural Language to API