paper-with-me

Papers

SafeCoT: Improving VLM Safety with Minimal Reasoning

2025-06-10 · Jiachen Ma, Zhanhui Zhou, Chao Yang, Chaochao Lu

Ensuring safe and appropriate responses from vision-language models (VLMs) remains a critical challenge, particularly in high-risk or ambiguous scenarios. We introduce SafeCoT, a lightweight, interpretable framework that leverages rule-based chain-of-thought (CoT) supervision to improve refusal behavior in VLMs. Unlike prior methods that rely on large-scale safety annotations or complex modeling, SafeCoT uses minimal supervision to help models reason about safety risks and make context-aware refusals. Experiments across multiple benchmarks show that SafeCoT significantly reduces overrefusal and enhances generalization, even with limited training data. Our approach offers a scalable solution for aligning VLMs with safety-critical objectives.

📄 PDF Abstract BibTeX arXiv:2506.08399

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments

2025-08-06 · Yuquan Wang, Mi Zhang, Yining Wang, Geng Hong 외 arxiv

Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning proc…

AlphaAlign: Incentivizing Safety Alignment with Extremely Simplified Reinforcement Learning

2025-07-20 · Yi Zhang, An Zhang, XiuYu Zhang, Leheng Sheng 외 arxiv

Large language models (LLMs), despite possessing latent safety understanding from their vast pretraining data, remain vulnerable to generating harmful content and exhibit issues such as over-refusal and utility degradati…

Reinforcement Learning

Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model

2026-04-01 · Pankayaraj Pathmanathan, Furong Huang arxiv

While the wide adoption of refusal training in large language models (LLMs) has showcased improvements in model safety, recent works have highlighted shortcomings due to the shallow nature of these alignment methods. To …

MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models

2026-04-17 · Manh Luong, Tamas Abraham, Junae Kim, Amar Kaur 외 arxiv

Existing multimodal safety benchmarks focus solely on visual inputs and cannot assess Omni Large Language Models (LLMs) that process vision, audio, and text. We introduce MCBench, a benchmark with 1196 scenarios spanning…

Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Training

2025-10-23 · Zheng-Xin Yong, Stephen H. Bach arxiv

We discover a novel and surprising phenomenon of unintentional misalignment in reasoning language models (RLMs), which we call self-jailbreaking. Specifically, after benign reasoning training on math or code domains, RLM…