UnsafeChain: Enhancing Reasoning Model Safety via Hard Cases
This addresses safety issues in reasoning models for AI developers and users, but it is incremental as it builds on existing safety alignment methods.
The paper tackles the safety challenge in large reasoning models by introducing UnsafeChain, a dataset of hard prompts with unsafe completions corrected to safe responses, which when used for fine-tuning outperforms prior safety alignment datasets across multiple benchmarks, with a 1K subset matching or surpassing baseline performance.
As large reasoning models (LRMs) grow more capable, chain-of-thought (CoT) reasoning introduces new safety challenges. Existing SFT-based safety alignment studies dominantly focused on filtering prompts with safe, high-quality responses, while overlooking hard prompts that always elicit harmful outputs. To fill this gap, we introduce UnsafeChain, a safety alignment dataset constructed from hard prompts with diverse sources, where unsafe completions are identified and explicitly corrected into safe responses. By exposing models to unsafe behaviors and guiding their correction, UnsafeChain enhances safety while preserving general reasoning ability. We fine-tune three LRMs on UnsafeChain and compare them against recent SafeChain and STAR-1 across six out-of-distribution and five in-distribution benchmarks. UnsafeChain consistently outperforms prior datasets, with even a 1K subset matching or surpassing baseline performance, demonstrating the effectiveness and generalizability of correction-based supervision. We release our dataset and code at https://github.com/mbzuai-nlp/UnsafeChain