Activation Steering

Follow this topic Ask about this topic

Activation steering studies how edits to a model’s hidden activations can control its behavior at inference time, enabling changes without retraining the full model. Common approaches add or remove contrastive steering vectors, target sparse autoencoder features, neurons or attention heads, and use conditional or feedback-based controllers to adapt interventions to the input or generation state. Researchers apply these methods to language tasks such as reasoning, refusal, factuality and hallucination mitigation, as well as image and video generation, speech, and embodied agents; evaluations include benchmarks such as GSM8K, TruthfulQA, POPE and LIBERO. A central challenge is that readable or decodable features are not necessarily controllable, and effective edits can vary by context and architecture or cause collateral changes to fluency, utility and safety.

Latest papers

376 papers77 papers in the last four weeks, up 93% on the four weeks before. 0.5% of all new papers.

Evaluating the Semantic Specificity of Representation Steering in Language Models
Aug 29, 2026Zhangdie Yuan, Andreas VlachosLLM Interpretability Activation Steering
University of Cambridge
Attribute-Based Activation Steering of LLMs for Group-Specific Explanation Generation
Aug 29, 2026Leandra Fichtel, Janek Prange, Henning WachsmuthLanguage Model Steering Activation Steering
Leibniz University Hannover · Leibniz University Hannover, L3S Research Center
The Halt Vector: Internalizing a Causal Steering Intervention for Efficient Reasoning
Aug 28, 2026Dylan Jayabahu, Tinuade AdelekeEfficient Language Model Reasoning Overthinking in Language Models
University of Waterloo · Independent Researcher
Sycophancy Suppression Can Impair Rational Updating: Anti-Sycophancy Should Preserve the Ability to Update
Aug 27, 2026Huanhuan Ma, Henry Peng Zou, Chengze Li +3LLM Sycophancy Belief Updating in LLMs
University of Illinois Chicago · National University of Singapore
Controllable Affective Generation via Latent Vector Steering
Aug 26, 2026Xixian Yong, Siyuan Chang, Yingying Zhang +2Controllable Text Generation Activation Steering
Gaoling School of Artificial Intelligence, Renmin University of China · Tencent Jarvis Lab · Beijing Key Laboratory of Research on Large Models and Intelligent Governance +1
Reflection Steering: Disentangling Reflection from Reasoning in Activation Space for Token-Efficient Inference
Aug 26, 2026Jiarui Hu, Zhiyuan Wen, Xiaoyun Liu +2Language Model Steering Efficient Language Model Reasoning
School of Computing and Data Science, The University of Hong Kong, Hong Kong SAR, China · Department of Computing, The Hong Kong Polytechnic University, Hong Kong SAR, China · School of Data Science, Lingnan University, Hong Kong SAR, China +1
GGSS: Geodesic-Gated Spherical Steering for Inference-Time Debiasing of Generative Vision-Language Models
Aug 26, 2026Yiqun Sun, Junyu Chen, Pengfei Wei +1Activation Steering Counterfactual Evaluation of VLMs
Magellan Technology Research Institute (MTRI) · National University of Singapore
SteerCheck: Attribution Specificity and Alignment Leakage in Activation-Steering Audits
Aug 25, 2026Daming Luo, Christy Liang, Junyu XuanActivation Steering LLM Auditing
University of Technology Sydney
Semantic Overlays: Mitigating Prompt Injection with Annotations Beyond Tokens and Steering Vectors
Aug 24, 2026Joshua PenmanLLM Security Prompt Injection Defense
Gated Activation Steering for Reducing Sycophancy & Hallucination in Medical Question Answering
Aug 24, 2026Himanshu Tripathi, Subash Neupane, Shaswata Mitra +3LLM Sycophancy LLM Hallucination Mitigation
Department of Computer Science, The University of Alabama, Tuscaloosa, AL, USA
What Does Activation Steering Control? Attribution Across Answer Encodings and Output-Sensitive Subspaces
Aug 24, 2026Zhiwei Gao, Shaowen Peng, Shoko Wakamiya +1Activation Steering Transformer Interpretability
Nara Institute of Science and Technology
Encoded but Not Actionable: Auditing the Decode-Generate-Steer Gap in Frozen LLMs for Geometric Constraints
Aug 18, 2026Man Liang, Xinzhao Cheng, Faizan WajidLLM Auditing LLM Interpretability
University of Maryland
PCA-guided Activation Scaling for Monotonic Bidirectional Control over LLM Sycophancy
Aug 17, 2026Zheng Chen, Zhaoxin Feng, Yip Tin Po +3LLM Sycophancy Activation Steering
The Hong Kong University of Science and Technology · The Hong Kong Polytechnic University
THESIS-MoE: Trainable Hierarchical Extraction and SteerIng of Sycophancy in Mixture-of-Experts
Aug 16, 2026Kareem Hassani, Chaymaa Abbas, Lama Mawlawi +1LLM Sycophancy Activation Steering
Department of Electrical and Computer Engineering, American University of Beirut, Riad El-Solh, Beirut 1107 2020, Lebanon
Interpretable Cross-Lingual Alignment in Small Language Models: Probing Cultural and Pragmatic Reasoning in Japanese-English Bilingual LLMs
Aug 14, 2026Florian BraunPragmatic Reasoning in Language Models Cross-Lingual Reasoning in Language Models
Independent Researcher
You Only Pass Once: Answering and Abstaining Together in a Single Forward Pass of a Frozen Language Model
Aug 14, 2026Ziyang Luo, Zhongyao Chu, Xinjie He +4LLM Abstention Selective Prediction
Predictive Memory Localization: Forecasting Selective Intervention Paths from Internal Signals
Aug 13, 2026Jinhao Jing, Tian Zeyu, Lucas Qingyang Fang +4Selective Prediction Mechanistic Interpretability
The Chinese University of Hong Kong, Shenzhen · Shanghai Jiao Tong University · University of California, Santa Cruz +3
Semantic Steering for Controllable Generation: Tuning-Free Concept Erasure in Multimodal Diffusion Transformers
Aug 13, 2026Qiao Li, Xiaomeng Fu, Yuanshu Zhao +3Diffusion Model Unlearning Activation Steering
Institute of Information Engineering, Chinese Academy of Sciences Beijing, China · School of Cyber Security, University of Chinese Academy of Sciences Beijing, China
Located but Not Releasable: Silent Gate Inversion and Bounded Linear Release
Aug 12, 2026Xining XunCausal Interventions in Language Models Activation Steering
Tsingjiao Information Science (Beijing) Co., Ltd.
Probing and steering biology across Boltz-1s trunk-diffusion boundary
Aug 11, 2026Piotr Jedryszek, Tongmeng Xie, Adam Winnifrith +5Protein Structure Prediction Activation Steering
Department of Biology, University of Oxford, Oxford, UK · Evolvere Biosciences, London, UK · Independent Researcher +4
Your LLM, Your Style: Behavioral Mode Axes for LLM Behavioral Control
Aug 11, 2026Haoze Liu, Run Liu, Haiying Xu +6Personality Modeling in Language Models Activation Steering
Shanghai Jiao Tong University · Shanghai AI Laboratory · The Hong Kong University of Science and Technology +1
Multimodal Model Diffing for Feature Discovery and Control
Aug 10, 2026Hunar Batra, Lachin Naghashyar, Ashkan Khakzar +4Sparse Autoencoders Multimodal Model Interpretability
University of Oxford · Microsoft
Deployable Per-Instance Multi-Layer Activation Steering for Large Language Models
Aug 9, 2026Muhammad Faishal Adly Nelwan, Alfan Farizki WicaksonoLanguage Model Steering Controllable Text Generation
Faculty of Computer Science Universitas Indonesia
MI-MIDI: Mechanistic Interpretability of Text-to-MIDI Generation Models via Probing, Lenses and Steering
Aug 6, 2026Jakub Poćwiardowski, Mateusz ModrzejewskiControllable Music Generation Symbolic Music Generation
Institute of Computer Science Warsaw University of Technology Warsaw, Poland
From Behavior to Mechanism: Tracing Divergent Response Modes in Frontier Language Models
Aug 6, 2026Ali Jalal-KamaliLanguage Model Steering LLM Interpretability
USC Institute for Creative Technologies
CircuitSteer: Geometrically Aligned Multi-Layer Steering via Sparse Autoencoder Circuits
Aug 6, 2026Mehrshad Saadatinia, Parsa Razmara, Ardalan Aryashad +2LLM Alignment Language Model Steering
University of Southern California, Los Angeles, USA
FOCUS: Decoupling Expert Personas in LLMs to Enhance Domain Expert Capabilities
Aug 6, 2026Guanyu Wang, Zidi Zhang, Xu ChuLanguage Model Steering Personality Modeling in Language Models
Peking University, Beijing, China · The University of Sydney, Sydney, Australia
Latent Fact-Checking: Detecting Misinformation through Activation Engineering
Aug 5, 2026Pedro T. Barcelos, Otávio Parraga, Marcelo M. Mussi +3Fake News Detection Automated Fact-Checking
MALTA, Machine Learning Theory and Applications Lab, PUCRS, Porto Alegre, Brazil · Kunumi Institute, Brazil
Inverted Detection and Control in Steering Vectors
Aug 3, 2026Max Torop, Aria Masoomi, Jennifer DyLanguage Model Steering Causal Interventions in Language Models
Northeastern University
Through the LENS: Local Geometric Decomposition of Vision-Language Model Representations
Aug 1, 2026Shalom Kachko, Raz Lapid, Margarita Vald +2Vision-Language Models Representation Geometry in VLMs
1Ben-Gurion University of the Negev · 2Intuit Inc. · 3Deepkeep

Aug 29, 2026J/K move · Enter open · S save

Evaluating the Semantic Specificity of Representation Steering in Language Models

Zhangdie Yuan, Andreas Vlachos

University of Cambridge

LLM Interpretability Activation Steering LLM Reasoning Reasoning Consistency in Language Models

Open paper AskPDF arXiv

Localized Representation Steering (LRS) is widely used to correct reasoning pathologies in large language models. However, standard benchmark evaluations can easily be fooled by superficial label overrides, creating a false impression of reasoning circuit repairs. In this work, we propose Cross-Rule Transfer (CRT), a diagnostic framework that audits representational interventions by evaluating them on rule families where the model is natively competent. Evaluating late-layer LRS for a widespread logical failure, contradiction blindness, reveals that the intervention merely injects a global label bias: applying the steering vector to rules the model already handles correctly (99.6% baseline) degrades performance to 40.4% by forcing false contradiction predictions. We support this diagnosis with four complementary controls (direct logit bias equivalence, control vector label-flipping, cross-model grafting, and early-layer steering checks), providing a rigorous methodology to distinguish genuine reasoning repairs from superficial label overrides.