DriveSafe: A Hierarchical Risk Taxonomy for Safety-Critical LLM-Based Driving Assistants

📄 Abstract

Large Language Models (LLMs) are increasingly integrated into vehicle-based digital assistants, where unsafe, ambiguous, or legally incorrect responses can lead to serious safety, ethical, and regulatory consequences. Despite growing interest in LLM safety, existing taxonomies and evaluation frameworks remain largely general-purpose and fail to capture the domain-specific risks inherent to real-world driving scenarios. In this paper, we introduce DriveSafe, a hierarchical, four-level risk taxonomy designed to systematically characterize safety-critical failure modes of LLM-based driving assistants. The taxonomy comprises 129 fine-grained atomic risk categories spanning technical, legal, societal, and ethical dimensions, grounded in real-world driving regulations and safety principles and reviewed by domain experts. To validate the safety relevance and realism of the constructed prompts, we evaluate their refusal behavior across six widely deployed LLMs. Our analysis shows that the evaluated models often fail to appropriately refuse unsafe or non-compliant driving-related queries, underscoring the limitations of general-purpose safety alignment in driving contexts.

🔍 Key Points

Introduction of DriveSafe, a hierarchical four-level risk taxonomy for LLM-based driving assistants that includes 129 atomic risk categories.
Characterization of safety-critical failure modes in LLMs within specific driving contexts, highlighting risks that existing frameworks overlook.
Empirical analysis showing that current LLMs often fail to appropriately refuse unsafe or ambiguous driving-related prompts, demonstrating the inadequacy of general-purpose AI safety checks.
The use of expert review to validate the taxonomy ensures it aligns with real-world regulations and safety principles, adding credibility to its applicability.

💡 Why This Paper Matters

This paper is crucial as it lays the foundation for a structured approach to assessing the safety risks associated with LLM-based driving assistants, which are increasingly used in autonomous vehicles. By highlighting specific risk categories and failure modes, it aids in improving the safety and reliability of AI systems in high-stakes driving contexts, ultimately contributing to safer road environments and regulatory compliance.

🎯 Why It's Interesting for AI Security Researchers

AI security researchers would find this paper valuable as it addresses critical risks related to the deployment of LLMs in safety-sensitive applications. The detailed taxonomy provides a framework for identifying and mitigating risks that could lead to serious safety incidents. Furthermore, the demonstrated failures of existing models in recognizing and refusing unsafe prompts emphasize the need for improved alignment methods, which is an essential focus area in the field of AI security.

DriveSafe: A Hierarchical Risk Taxonomy for Safety-Critical LLM-Based Driving Assistants

📄 Abstract

🔍 Key Points

💡 Why This Paper Matters

🎯 Why It's Interesting for AI Security Researchers

📚 Read the Full Paper