GUARD: Guideline Upholding Test through Adaptive Role-play and Jailbreak Diagnostics for LLMs

2025-08-27

AI Summary

This work addresses the challenge of operationalizing governmental AI ethics guidelines into quantifiable safety evaluations. We propose the first automated framework that systematically translates high-level ethical principles into structured test cases: (1) constructing a multidimensional violation question bank via rule-based and generative methods; (2) simulating jailbreaking scenarios through adversarial role-playing and dynamic response analysis; and (3) introducing GUARD-JD, the first jailbreak diagnostic mechanism for detecting latent ethical violations. The framework supports cross-modal transfer testing for both text- and vision-language models. Evaluated on seven mainstream large language and multimodal models, it demonstrates strong efficacy in identifying both explicit and implicit ethical violations, generating fine-grained compliance reports, and significantly enhancing the operationalizability and depth of AI ethics assessments.

Technology Category

• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

As Large Language Models become increasingly integral to various domains, their potential to generate harmful responses has prompted significant societal and regulatory concerns. In response, governments have issued ethics guidelines to promote the development of trustworthy AI. However, these guidelines are typically high-level demands for developers and testers, leaving a gap in translating them into actionable testing questions to verify LLM compliance. To address this challenge, we introduce GUARD ( extbf{G}uideline extbf{U}pholding Test through extbf{A}daptive extbf{R}ole-play and Jailbreak extbf{D}iagnostics), a testing method designed to operationalize guidelines into specific guideline-violating questions that assess LLM adherence. To implement this, GUARD uses automated generation of guideline-violating questions based on government-issued guidelines, thereby testing whether responses comply with these guidelines. When responses directly violate guidelines, GUARD reports inconsistencies. Furthermore, for responses that do not directly violate guidelines, GUARD integrates the concept of ``jailbreaks'' to diagnostics, named GUARD-JD, which creates scenarios that provoke unethical or guideline-violating responses, effectively identifying potential scenarios that could bypass built-in safety mechanisms. Our method finally culminates in a compliance report, delineating the extent of adherence and highlighting any violations. We have empirically validated the effectiveness of GUARD on seven LLMs, including Vicuna-13B, LongChat-7B, Llama2-7B, Llama-3-8B, GPT-3.5, GPT-4, GPT-4o, and Claude-3.7, by testing compliance under three government-issued guidelines and conducting jailbreak diagnostics. Additionally, GUARD-JD can transfer jailbreak diagnostics to vision-language models, demonstrating its usage in promoting reliable LLM-based applications.

Problem

• Translating high-level AI ethics guidelines into actionable testing questions
• Assessing LLM compliance with government-issued ethical guidelines
• Identifying potential scenarios that bypass built-in safety mechanisms

Innovation

• Automated generation of guideline-violating questions
• Integration of jailbreak diagnostics for safety testing
• Compliance reporting with adherence and violation delineation