English
Related papers

Related papers: One-shot Optimized Steering Vectors Mediate Safety…

200 papers

The safety mechanisms of large language models (LLMs) exhibit notable fragility, as even fine-tuning on datasets without harmful content may still undermine their safety capabilities. Meanwhile, existing safety alignment methods…

Computers and Society · Computer Science 2026-02-03 Guanghao Zhou , Panjia Qiu , Cen Chen , Hongyu Li , Mingyuan Chu , Xin Zhang , Jun Zhou

Alignment of Large Language Models (LLMs) is the ability to satisfy desired objectives during generation, which is critical for trustworthy deployment. In practice, alignment is often operationalized through multiple objectives such as…

Computation and Language · Computer Science 2026-05-19 Gautam Siddharth Kashyap , Mark Dras , Usman Naseem

Large language models (LLMs) can sometimes detect when they are being evaluated and adjust their behavior to appear more aligned, compromising the reliability of safety evaluations. In this paper, we show that adding a steering vector to an…

Computation and Language · Computer Science 2026-03-03 Tim Tian Hua , Andrew Qin , Samuel Marks , Neel Nanda

As the capabilities of Vision Language Models (VLMs) continue to improve, they are increasingly targeted by jailbreak attacks. Existing defense methods face two major limitations: (1) they struggle to ensure safety without compromising the…

Cryptography and Security · Computer Science 2025-09-29 Xiyu Zeng , Siyuan Liang , Liming Lu , Haotian Zhu , Enguang Liu , Jisheng Dang , Yongbin Zhou , Shuchao Pang

Precise control over language model generation is vital for ensuring both safety and reliability. Although prompt engineering and steering are commonly used to intervene in model behaviors, the vast number of parameters in models often…

Computation and Language · Computer Science 2025-06-04 Mengru Wang , Ziwen Xu , Shengyu Mao , Shumin Deng , Zhaopeng Tu , Huajun Chen , Ningyu Zhang

Large Language Models (LLMs) have emerged as powerful tools, but their inherent safety risks - ranging from harmful content generation to broader societal harms - pose significant challenges. These risks can be amplified by the recent…

Recent vision language models (VLMs) have made remarkable strides in generative modeling with multimodal inputs, particularly text and images. However, their susceptibility to generating harmful content when exposed to unsafe queries raises…

Artificial Intelligence · Computer Science 2026-03-06 Yiwei Chen , Yuguang Yao , Yihua Zhang , Bingquan Shen , Gaowen Liu , Sijia Liu

Multi-modal large language models (MLLMs) have made significant progress, yet their safety alignment remains limited. Typically, current open-source MLLMs rely on the alignment inherited from their language module to avoid harmful…

Cryptography and Security · Computer Science 2025-04-15 Yanbo Wang , Jiyang Guan , Jian Liang , Ran He

We show that reinforcement learning with verifiable reward using one training example (1-shot RLVR) is effective in incentivizing the math reasoning capabilities of large language models (LLMs). Applying RLVR to the base model…

Activation steering is a method for controlling Large Language Model (LLM) behavior by intervening in its internal representations to increase the alignment with a specific target feature direction. However, standard interventions, such as…

Machine Learning · Computer Science 2026-05-05 Tam Nguyen , Tu Anh Nguyen , Sina Alemohammad , Richard G. Baraniuk

While large language models (LLMs) are trained to align with human values, their generations may still violate safety constraints. A growing line of work addresses this problem by modifying the model's sampling policy at decoding time using…

Machine Learning · Computer Science 2026-05-15 Bat-Sheva Einbinder , Hen Davidov , Yee Whye Teh , Yarin Gal , Yaniv Romano

Recent work has shown that LLMs can sometimes detect when steering vectors are injected into their residual stream and identify the injected concept -- a phenomenon termed "introspective awareness." We investigate the mechanisms underlying…

Machine Learning · Computer Science 2026-05-18 Uzay Macar , Li Yang , Atticus Wang , Peter Wallich , Emmanuel Ameisen , Jack Lindsey

Vision Language Models (VLMs) have shown remarkable performance, but are also vulnerable to backdoor attacks whereby the adversary can manipulate the model's outputs through hidden triggers. Prior attacks primarily rely on single-modality…

Computer Vision and Pattern Recognition · Computer Science 2025-06-10 Zhiyuan Zhong , Zhen Sun , Yepang Liu , Xinlei He , Guanhong Tao

Modern large language models (LLMs) are typically secured by auditing data, prompts, and refusal policies, while treating the forward pass as an implementation detail. We show that intermediate activations in decoder-only LLMs form a…

Cryptography and Security · Computer Science 2025-11-24 Zhiyuan Xu , Stanislav Abaimov , Joseph Gardiner , Sana Belguith

One way to mitigate risks in vision-language models (VLMs) is to remove dangerous samples in their training data. However, such data moderation can be easily bypassed when harmful images are split into small, benign-looking patches,…

Computer Vision and Pattern Recognition · Computer Science 2025-06-06 Zhanhui Zhou , Lingjie Chen , Chao Yang , Chaochao Lu

MLLMs are increasingly deployed in multi-turn settings, where attackers can escalate unsafe intent through the evolving visual-text history and exploit long-context safety decay. Yet safety alignment is still dominated by single-turn data…

Machine Learning · Computer Science 2026-05-28 Haolong Hu , Hanyu Li , Tiancheng He , Huahui Yi , An Zhang , Qiankun Li , Kun Wang , Yang Liu , Zhigang Zeng

Recent advancements in Reinforcement Learning with Verifiable Rewards (RLVR) have gained significant attention due to their objective and verifiable reward signals, demonstrating strong performance in reasoning and code generation tasks.…

Cryptography and Security · Computer Science 2025-10-20 Yuexiao Liu , Lijun Li , Xingjun Wang , Jing Shao

Vision-Language adaptation (VL adaptation) transforms Large Language Models (LLMs) into Large Vision-Language Models (LVLMs) for multimodal tasks, but this process often compromises the inherent safety capabilities embedded in the original…

Computation and Language · Computer Science 2024-11-18 Seongyun Lee , Geewook Kim , Jiyeon Kim , Hyunji Lee , Hoyeon Chang , Sue Hyun Park , Minjoon Seo

Vision-language models (VLMs) are increasingly deployed in real-world and embodied settings where safety decisions depend on visual context. However, it remains unclear which visual evidence drives these judgments. We study whether…

Computer Vision and Pattern Recognition · Computer Science 2026-03-20 Carlos Hinojosa , Clemens Grange , Bernard Ghanem

Recent advances in large language models (LLMs) have led to the development of thinking language models that generate extensive internal reasoning chains before producing responses. While these models achieve improved performance,…

Machine Learning · Computer Science 2025-10-23 Constantin Venhoff , Iván Arcuschin , Philip Torr , Arthur Conmy , Neel Nanda