StealthGraph: Exposing Domain-Specific Risks in LLMs through Knowledge-Graph-Guided Harmful Prompt Generation

2026-01-08

arXiv.org

AI Summary

This work addresses the scarcity of implicitly harmful prompts in domain-specific large language model safety evaluations, a limitation exacerbated by the inability of manually crafted prompts to reflect real-world threats. To overcome this, the authors propose a novel approach that integrates knowledge graph–guided generation with dual-path obfuscation rewriting. Specifically, domain-relevant prompts are first generated using a knowledge graph, then transformed into highly implicit harmful forms through two complementary strategies: direct rewriting and context-aware enhancement. This methodology enables the first systematic construction of a red-teaming dataset that simultaneously exhibits strong domain relevance and high levels of implicitness. Empirical results demonstrate that the resulting dataset significantly improves the effectiveness of safety evaluations for domain-specific models. The authors publicly release both the code and the dataset to support future research in this area.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs) are increasingly applied in specialized domains such as finance and healthcare, where they introduce unique safety risks. Domain-specific datasets of harmful prompts remain scarce and still largely rely on manual construction; public datasets mainly focus on explicit harmful prompts, which modern LLM defenses can often detect and refuse. In contrast, implicit harmful prompts-expressed through indirect domain knowledge-are harder to detect and better reflect real-world threats. We identify two challenges: transforming domain knowledge into actionable constraints and increasing the implicitness of generated harmful prompts. To address them, we propose an end-to-end framework that first performs knowledge-graph-guided harmful prompt generation to systematically produce domain-relevant prompts, and then applies dual-path obfuscation rewriting to convert explicit harmful prompts into implicit variants via direct and context-enhanced rewriting. This framework yields high-quality datasets combining strong domain relevance with implicitness, enabling more realistic red-teaming and advancing LLM safety research. We release our code and datasets at GitHub.

Problem

• domain-specific risks
• implicit harmful prompts
• large language models
• knowledge graph
• LLM safety

Innovation

• knowledge-graph-guided generation
• implicit harmful prompts
• dual-path obfuscation
• domain-specific red-teaming
• LLM safety