SafeRBench: A Comprehensive Benchmark for Safety Assessment in Large Reasoning Models
This addresses safety assessment for users and developers of LRMs, but it is incremental as it builds on existing safety evaluation methods by adding process-level analysis.
The authors tackled the problem of safety risks in Large Reasoning Models (LRMs) by introducing SafeRBench, a benchmark for end-to-end safety assessment, and demonstrated its effectiveness through evaluations on 19 LRMs, providing detailed insights into risks and protective mechanisms.
Large Reasoning Models (LRMs) improve answer quality through explicit chain-of-thought, yet this very capability introduces new safety risks: harmful content can be subtly injected, surface gradually, or be justified by misleading rationales within the reasoning trace. Existing safety evaluations, however, primarily focus on output-level judgments and rarely capture these dynamic risks along the reasoning process. In this paper, we present SafeRBench, the first benchmark that assesses LRM safety end-to-end -- from inputs and intermediate reasoning to final outputs. (1) Input Characterization: We pioneer the incorporation of risk categories and levels into input design, explicitly accounting for affected groups and severity, and thereby establish a balanced prompt suite reflecting diverse harm gradients. (2) Fine-Grained Output Analysis: We introduce a micro-thought chunking mechanism to segment long reasoning traces into semantically coherent units, enabling fine-grained evaluation across ten safety dimensions. (3) Human Safety Alignment: We validate LLM-based evaluations against human annotations specifically designed to capture safety judgments. Evaluations on 19 LRMs demonstrate that SafeRBench enables detailed, multidimensional safety assessment, offering insights into risks and protective mechanisms from multiple perspectives.