Safety alignment incurs safety tax that perturbs a large reasoning model's (LRM) general reasoning ability. Existing datasets used for safety alignment for an LRM are usually constructed by distilling safety reasoning traces and answers from an external LRM or human labeler. However, such reasoning traces and answers exhibit a distributional gap with the target LRM that needs alignment, and we conjecture such distributional gap is the culprit leading to significant degradation of reasoning ability of the target LRM. Driven by this hypothesis, we propose a safety alignment dataset construction method, dubbed DGR. DGR transforms and refines an existing out-of-distributional safety reasoning dataset to be aligned with the target's LLM inner distribution. Experimental results demonstrate that i) DGR effectively mitigates the safety tax while maintaining safety performance across all baselines, i.e., achieving \textbf{+30.2\%} on DirectRefusal and \textbf{+21.2\%} on R1-ACT improvement in average reasoning accuracy compared to Vanilla SFT; ii) the degree of reasoning degradation correlates with the extent of distribution shift, suggesting that bridging this gap is central to preserving capabilities. Furthermore, we find that safety alignment in LRMs may primarily function as a mechanism to activate latent knowledge, as a mere \textbf{10} samples are sufficient for activating effective refusal behaviors. These findings not only emphasize the importance of distributional consistency but also provide insights into the activation mechanism of safety in reasoning models.
Using Grounded Theory for Agent Behavior Analysis at Scale
Purdue University(普渡大学);Stevens Institute of Technology(史蒂文斯理工学院);Cornell University(康奈尔大学);University of Texas at El Paso(德克萨斯大学埃尔帕索分校);Johns Hopkins University(约翰斯·霍普金斯大学)
Mitigating LLM Hallucinations through Domain-Grounded Tiered Retrieval
Computational Unit, Z.H. College of Engineering & Technology, Aligarh Muslim University, India(计算单元,Z.H. 工程技术学院,阿利加尔穆斯林大学,印度);Interdisciplinary Center for Artificial Intelligence, Aligarh Muslim University, India(跨学科人工智能中心,阿利加尔穆斯林大学,印度);Department of Computer Engineering, Aligarh Muslim University, India(计算机工程系,阿利加尔穆斯林大学,印度)
Safety Testing LLM Agents at Scale: From Risk Discovery to Evidence-Grounded Verification
AntGroup(蚂蚁集团);Zhejiang University(浙江大学);Fudan University(复旦大学);Alibaba Group(阿里巴巴集团);Hunan Institute of Advanced Technology(湖南先进技术研究院);Deakin University(迪肯大学)
HalluClear: Diagnosing, Evaluating and Mitigating Hallucinations in GUI Agents
MAIS&NLPR, Institute of Automation, Chinese Academy of Sciences(自动化研究所人工智能与自然语言处理实验室,中国科学院);School of Artificial Intelligence, UCAS(中国科学院大学人工智能学院);Meituan(美团)
Beyond Blind Spots: Analytic Hints for Mitigating LLM-Based Evaluation Pitfalls
Ora Nova Fandina(奥拉诺瓦·法丁纳);Eitan Farchi(埃itan·法奇);Shmulik Froimovich(什mulik·弗罗伊米奇);Raviv Gal(拉维夫·加尔);Wesam Ibraheem(韦萨姆·伊布拉希姆);Rami Katan(拉米·卡坦);Alice Podolsky(艾丽斯·波德洛斯基)
SAVE: Sparse Autoencoder-Driven Visual Information Enhancement for Mitigating Object Hallucination
Department of Electrical and Computer Engineering, Seoul National University(电子与计算机工程系,首尔国立大学);Daegu Gyeongbuk Institute of Science and Technology(大邱庆州科学技术院);IPAI, AIIS, ASRI, INMC, and ISRC, Seoul National University(IPAI、AIIS、ASRI、INMC 和 ISRC,首尔国立大学)
Quantifying and Mitigating Self-Preference Bias of LLM Judges
CompleX Lab, School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, China(复杂实验室、计算机科学与工程学院、电子科技大学、成都、中国)