Representation Engineering

Follow this topic Ask about this topic

Representation engineering studies the internal activations of language, vision, audio, and multimodal models, asking how concepts and behaviors are encoded and how interventions can change them. Common approaches identify directions or subspaces using contrastive examples, principal component analysis, linear probes, or sparse autoencoders, then steer activations at inference time or edit model components. These methods are used for safety and refusal, hallucination reduction, preference alignment, reasoning, and adaptation across domains or languages. A central challenge is making interventions reliable while avoiding collateral changes to fluency, accuracy, routing, or other capabilities, and evaluation must test both the intended effect and those side effects.

Latest papers

23 papers2 papers in the last four weeks, down 33% on the four weeks before. 0.0% of all new papers.

Mitigating Reward Hacking in Realistic Scenarios Through Representation Engineering
Oct 21, 2026Austin Meek, Kyle Cox, Matthew Nguyen +2Reward Hacking Activation Steering
University of Delaware · Independent
Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs
Oct 19, 2026Zijian Liu, Chaoyou Fu, Yan Yang +2LLM Hallucination Mitigation Multimodal Hallucination
School of Intelligence Science and Technology, Nanjing University, Suzhou 215163, China. · University of Chinese Academy of Sciences, Beijing 100049, China.
Representation Editing for Multimodal Test-Time Adaptation
Sep 26, 2026Longfei Huang, Xiangyu Wu, Yang YangTest-Time Adaptation Multimodal Robustness
Nanjing University of Science and Technology · Alibaba Group
Bait-and-Recover: Poisoning Internal Refusal Signals to Defend LLMs against White-Box Editing Jailbreaks
Sep 5, 2026Tian Gao, Zhipeng Xie, Yuhao Wu +2LLM Security Adversarial Attacks on LLMs
Anhui Laboratory for Safe Artificial Intelligence in the Yangtze River Delta · iFlytek Research
Transfer Safety Awareness for Cross-Modal Safety Drift in Multimodal Large Language Models
Sep 2, 2026Tianqi Xiao, Shiyao Cui, Minghao Zhang +2Multimodal Multi-Turn Safety Alignment Multimodal Large Language Models
1The Conversational AI (CoAI) group, DCST, Tsinghua University · 3Northwestern Polytechnical University · School of Cyberspace Security, Beijing University of Posts and Telecommunications +1
Analyzing and Mitigating Cross-Lingual Degradation in Multilingual Medical VQA
Aug 23, 2026Jingbo Wang, Sendong Zhao, Haochun Wang +2Medical VQA Medical VLMs
Research Center for Social Computing and Interactive Robotics Harbin Institute of Technology, China
RARE: Decoupling Representation Steering from Expert Routing in Mixture-of-Experts Language Models
Aug 21, 2026Zhibo Zhang, Zhen Ouyang, Ling Shi +1Language Model Steering Representation Engineering
Huazhong University of Science and Technology Wuhan, China · AIDX TECH PTE. LTD. Singapore · Huazhong University of Science and Technology, Wuhan, China +1
Architectural Backdoors in Vision-Language Model Supply Chains via Representation Steering
Jul 28, 2026Maria Rosaria Briglia, Igor Maljkovic, Antonio Emanuele Cinà +3Adversarial Attacks on VLMs Representation Engineering
Sapienza University of Rome, Italy · University of Genoa, Italy · University of Trieste, Italy +1
LoRA for Gender-Inclusive Rewriting and Activation Steering for Counter-Narrative Generation
Jul 25, 2026Akhil Rajeev P, Manoj Balaji JGender Bias in Language Models Language Model Steering
C-DAC, Bangalore · Indian Institute of Technology, Kharagpur
Pulling The REINS: Training-Free Safety Alignment of Video Diffusion Models via Representation Steering
Jun 15, 2026Rohit Kundu, Arindam Dutta, Sarosij Bose +2Diffusion Model Alignment Video Diffusion Models
University of California, Riverside · YouTube (Google)
The Signs Were Always There: Training-Free Detection and Control in Raw Transformer Dimensions
Jun 10, 2026Varun Reddy Nalagatla, Madhavi Evana, Yiwen ZhangTransformer Interpretability Representation Engineering
Amazon Web Services
Whisper Hallucination Detection and Mitigation via Hidden Representation Steering and Sparse AutoEncoders
Jun 5, 2026Georgii Aparin, Vadim Popov, Tasnima Sadekova +1Audio-Language Model Hallucination Detection Sparse Autoencoders
AI Foundation and Algorithm Lab · National Research University Higher School of Economics
Towards the Readability of LLM-Generated Codes through Multitask Representation Engineering
Jun 4, 2026Huifan Gao, Liuhua He, Yinghui Pan +4Language Model Steering Representation Engineering
School of Aerospace Engineering, Xiamen University, Xiamen, China · School of Artificial Intelligence, Shenzhen University, Shenzhen, China · College of Computer and Cyber Security, Fujian Normal University, Fuzhou, China +3
HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models
Jun 2, 2026Shuang Liu, Yuxuan Bo, Qiuyang Zhao +4Reward Modeling Reward Hacking
Carnegie Mellon University · University of Virginia · Harvard University +4
REED: Post-Training Representation Editing for Cross-Domain Linguistic Steganalysis
May 27, 2026Ruohan Lei, Jianxin Gao, Wanli Peng +1Cross-Domain Generalization Representation Engineering
China Agricultural University, Beijing, China · Jiangsu Normal University, Xuzhou, China
SwordBench: Evaluating Orthogonality of Steering Image Representations
May 10, 2026Vladimir Zaigrajew, Dawid Pludowski, Hubert Baniecki +1Vision Foundation Models Representation Engineering
Centre for Credible AI Warsaw University of Technology Warsaw, Poland · University of Warsaw Warsaw, Poland
Knowledge Vector of Logical Reasoning in Large Language Models
Apr 26, 2026Zixuan Wang, Yuanyuan LeiLLM Interpretability Representation Engineering
Department of Computer & Information Science and Engineering University of Florida, Gainesville, FL, United States
Pref-CTRL: Preference Driven LLM Alignment using Representation Editing
Apr 26, 2026Imranul Ashrafi, Inigo Jauregi Unanue, Massimo PiccardiLLM Alignment Representation Engineering
University of Technology Sydney, Australia
Psychological Steering of Large Language Models
Apr 15, 2026Leonardo Blas, Robin Jia, Emilio FerraraLanguage Model Steering Representation Engineering
Thomas Lord Department of Computer Science, University of Southern California, Los Angeles, CA 90089
When Negation Is a Geometry Problem in Vision-Language Models
Mar 20, 2026Fawaz Sammani, Tzoulio Chamiti, Paul Gavrikov +1VLM Reasoning VLM Evaluation
ETRO Department, Vrije Universiteit Brussel, Belgium · imec, Kapeldreef 75, B-3001 Leuven, Belgium · Independent Researcher
Understanding and Defending VLM Jailbreaks via Jailbreak-Related Representation Shift
Mar 18, 2026Zhihua Wei, Qiang Li, Jian Ruan +6VLM Safety Adversarial Attacks on VLMs
School of Computer Science and Technology, Tongji University · Shanghai Artificial Intelligence Laboratory
HulluEdit: Single-Pass Evidence-Consistent Subspace Editing for Mitigating Hallucinations in Large Vision-Language Models
Feb 26, 2026Yangguang Lin, Quan Fang, Yufei Li +3VLM Hallucination Mitigation Object Hallucination in VLMs
Beijing University of Posts and Telecommunications, Beijing, China · Institute of Automation, Chinese Academy of Sciences, Beijing, China · Beijing Jiaotong University, Beijing, China
Beyond Forgetting: Representation Misdirection Elicits Controllable Side Behaviors and Capabilities
Jan 29, 2026Tien Dang, The-Hai Nguyen, Dinh Mai Phuong +5Linear Representation Hypothesis Controllable Text Generation
Japan Advanced Institute of Science and Technology · Monash University · Quantum AI and Cyber Security Institute, FPT Corporation +1

Oct 21, 2026J/K move · Enter open · S save

Mitigating Reward Hacking in Realistic Scenarios Through Representation Engineering

Austin Meek, Kyle Cox, Matthew Nguyen, Austin J. Brockmeier+1

University of Delaware · Independent

Reward Hacking Activation Steering Representation Engineering

Open paper AskPDF arXiv

Reward hacking is a major problem in current production machine learning systems, resulting in costly and dangerous failures. Here we study how to monitor for and mitigate reward hacking in realistic settings, including tool calls and long contexts. We apply a dynamic intervention to steering directions obtained by three different methods: difference-in-means vectors from labeled rollouts, gradient optimization-based vectors, and vectors extracted when models reference an anti-reward-hacking model spec. This dynamic intervention perturbs model representations less than uniform steering vector addition to the residual stream, allowing for more precise edits. We find that we are able to cheaply and significantly reduce reward hacking without correspondingly significant costs to model coherence or capability, and that steering vectors originally extracted from coding environments also generalize to reducing reward hacking in other domains. Our results pave the way towards latent space monitoring and control over realistic production systems.