SafeLawBench: Towards Safe Alignment of Large Language Models

2025-06-07

AI Summary

Current LLM safety evaluation lacks objective, reproducible benchmarks, with existing methods suffering from subjectivity and ill-defined criteria. To address this, we introduce SafeLawBench—the first legal-normative benchmark for LLM safety assessment—comprising 24,860 multiple-choice questions and 1,106 open-ended QA tasks. It features a novel three-tier legal risk taxonomy and establishes a multi-granular, verifiable safety alignment evaluation framework. We systematically evaluate 20 state-of-the-art models (including GPT-4o and Claude-3.5-Sonnet) using zero-shot/few-shot prompting, refusal behavior analysis, reasoning stability testing, and majority-voting ensembling. Results reveal that even top-performing models achieve only 80.5% accuracy, with an overall average of 68.8%, exposing a substantial gap in legal compliance. SafeLawBench provides a high-fidelity, reproducible, law-grounded evaluation paradigm for advancing LLM safety alignment research.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

With the growing prevalence of large language models (LLMs), the safety of LLMs has raised significant concerns. However, there is still a lack of definitive standards for evaluating their safety due to the subjective nature of current safety benchmarks. To address this gap, we conducted the first exploration of LLMs' safety evaluation from a legal perspective by proposing the SafeLawBench benchmark. SafeLawBench categorizes safety risks into three levels based on legal standards, providing a systematic and comprehensive framework for evaluation. It comprises 24,860 multi-choice questions and 1,106 open-domain question-answering (QA) tasks. Our evaluation included 2 closed-source LLMs and 18 open-source LLMs using zero-shot and few-shot prompting, highlighting the safety features of each model. We also evaluated the LLMs' safety-related reasoning stability and refusal behavior. Additionally, we found that a majority voting mechanism can enhance model performance. Notably, even leading SOTA models like Claude-3.5-Sonnet and GPT-4o have not exceeded 80.5% accuracy in multi-choice tasks on SafeLawBench, while the average accuracy of 20 LLMs remains at 68.8%. We urge the community to prioritize research on the safety of LLMs.

Problem

• Lack of definitive safety standards for LLM evaluation
• Subjective nature of current safety benchmarks
• Need for legal perspective in safety assessment

Innovation

• Proposes SafeLawBench benchmark for legal safety evaluation
• Includes 24,860 multi-choice and 1,106 open QA tasks
• Uses majority voting to enhance model performance