English
Related papers

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

200 papers

Large language models (LLMs) increasingly serve as automated evaluators, yet they suffer from "self-preference bias": a tendency to favor their own outputs over those of other models. This bias undermines fairness and reliability in…

Computation and Language · Computer Science 2025-09-05 Dani Roytburg , Matthew Bozoukov , Matthew Nguyen , Jou Barzdukas , Simon Fu , Narmeen Oozeer

As Vision Language Models (VLMs) are deployed across safety-critical applications, understanding and controlling their behavioral patterns has become increasingly important. Existing behavioral control methods face significant limitations:…

Computer Vision and Pattern Recognition · Computer Science 2025-10-01 Ravikumar Balakrishnan , Mansi Phute

Activation steering is a practical post-training model alignment technique to enhance the utility of Large Language Models (LLMs). Prior to deploying a model as a service, developers can steer a pre-trained model toward specific behavioral…

Cryptography and Security · Computer Science 2026-02-06 Chen Xiong , Zhiyuan He , Pin-Yu Chen , Ching-Yun Ko , Tsung-Yi Ho

Fine-tuning large language models (LLMs) to adapt to evolving safety policies is costly and impractical. Mechanistic interpretability enables inference-time control through latent activation steering, yet its potential for precise,…

Machine Learning · Computer Science 2025-06-06 Shaona Ghosh , Amrita Bhattacharjee , Yftah Ziser , Christopher Parisien

Hallucinations in LLMs pose a significant concern to their safe deployment in real-world applications. Recent approaches have leveraged the latent space of LLMs for hallucination detection, but their embeddings, optimized for linguistic…

Machine Learning · Computer Science 2025-05-23 Seongheon Park , Xuefeng Du , Min-Hsuan Yeh , Haobo Wang , Yixuan Li

Large language models (LLMs) have shown great potential as general-purpose AI assistants in various domains. To meet the requirements of different applications, LLMs are often customized by further fine-tuning. However, the powerful…

Machine Learning · Computer Science 2023-11-07 Xin Zhou , Yi Lu , Ruotian Ma , Tao Gui , Qi Zhang , Xuanjing Huang

Large Language Models (LLMs) have been extensively used across diverse domains, including virtual assistants, automated code generation, and scientific research. However, they remain vulnerable to jailbreak attacks, which manipulate the…

Cryptography and Security · Computer Science 2026-01-05 Haoran Gu , Handing Wang , Yi Mei , Mengjie Zhang , Yaochu Jin

Despite significant progress in alignment, large language models (LLMs) remain vulnerable to adversarial attacks that elicit harmful behaviors. Activation steering techniques offer a promising inference-time intervention approach, but…

Machine Learning · Computer Science 2026-01-28 Quy-Anh Dang , Chris Ngo

Latent space steering methods provide a practical approach to controlling large language models by applying steering vectors to intermediate activations, guiding outputs toward desired behaviors while avoiding retraining. Despite their…

Machine Learning · Computer Science 2026-01-13 Shawn Im , Sharon Li

Language models (LMs) are increasingly used in high-stakes, multi-agent settings, where following instructions and maintaining value alignment are critical. Most alignment research focuses on interactions between a single LM and a single…

Artificial Intelligence · Computer Science 2026-05-12 Maria Chang , Ronny Luss , Miao Liu , Keerthiram Murugesan , Karthikeyan Ramamurthy , Djallel Bouneffouf

Large language models (LLMs) have demonstrated impressive capabilities in natural language understanding and generation, but controlling their behavior reliably remains challenging, especially in open-ended generation settings. This paper…

Computation and Language · Computer Science 2025-12-08 Zirui He , Mingyu Jin , Bo Shen , Ali Payani , Yongfeng Zhang , Mengnan Du

Alignment in LLMs is more brittle than commonly assumed: misalignment can be triggered by adversarial prompts, benign fine-tuning, emergent misalignment, and goal misgeneralization. Recent evidence suggests that some misalignment behaviors…

Artificial Intelligence · Computer Science 2026-04-10 Niklas Herbster , Martin Zborowski , Alberto Tosato , Gauthier Gidel , Tommaso Tosato

Benefiting from the powerful capabilities of Large Language Models (LLMs), pre-trained visual encoder models connected to an LLMs can realize Vision Language Models (VLMs). However, existing research shows that the visual modality of VLMs…

Computer Vision and Pattern Recognition · Computer Science 2024-05-24 Zhendong Liu , Yuanbi Nie , Yingshui Tan , Xiangyu Yue , Qiushi Cui , Chongjun Wang , Xiaoyong Zhu , Bo Zheng

Safety prompts constitute an interpretable layer of defense against jailbreak attacks in vision-language models (VLMs); however, their efficacy is constrained by the models' latent structural responsiveness. We observe that such prompts…

Computer Vision and Pattern Recognition · Computer Science 2026-03-17 Chongxin Li , Hanzhang Wang , Lian Duan

Benefiting from the powerful capabilities of Large Language Models (LLMs), pre-trained visual encoder models connected to LLMs form Vision Language Models (VLMs). However, recent research shows that the visual modality in VLMs is highly…

Computer Vision and Pattern Recognition · Computer Science 2025-01-14 Zhendong Liu , Yuanbi Nie , Yingshui Tan , Jiaheng Liu , Xiangyu Yue , Qiushi Cui , Chongjun Wang , Xiaoyong Zhu , Bo Zheng

We introduce SteeringSafety, a systematic framework for evaluating representation steering methods across seven safety perspectives spanning 17 datasets. While prior work highlights general capabilities of representation steering, we…

Artificial Intelligence · Computer Science 2025-10-17 Vincent Siu , Nicholas Crispino , David Park , Nathan W. Henry , Zhun Wang , Yang Liu , Dawn Song , Chenguang Wang

The mechanisms by which reasoning training reshapes LLMs' internal computations remain unclear. We study lightweight steering vectors inserted into the base model's residual stream and trained with a reinforcement-learning objective. These…

The safety of large language models (LLMs) has increasingly emerged as a fundamental aspect of their development. Existing safety alignment for LLMs is predominantly achieved through post-training methods, which are computationally…

Artificial Intelligence · Computer Science 2026-02-03 Sicheng Shen , Mingyang Lv , Han Shen , Jialin Wu , Binghao Wang , Zhou Yang , Guobin Shen , Dongcheng Zhao , Feifei Zhao , Yi Zeng

Safety-aligned LLMs suffer from two failure modes: jailbreak (answering harmful inputs) and over-refusal (declining benign queries). Existing vector steering methods adjust the magnitude of answer vectors, but this creates a fundamental…

Machine Learning · Computer Science 2026-05-05 Haonan Zhang , Dongxia Wang , Yi Liu , Kexin Chen , Wenhai Wang

Activation steering methods enable inference-time control of large language model (LLM) behavior without retraining, but current approaches face a fundamental trade-off: sample-efficient methods suboptimally capture steering signals from…

Machine Learning · Computer Science 2026-03-09 Kartik Sharma , Rakshit S. Trivedi