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.

A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use
Jul 31, 2026Yutong Ke, Ming Yin, Chongwen Zhao +1LLM Interpretability LLM Tool Use
On the Generalization of Steering Vectors for Chain-of-Thought Faithfulness
Jul 31, 2026Matthew Nguyen, Kyle Cox, Austin Meek +1CoT Faithfulness Language Model Steering
University of Virginia · Independent · University of Delaware +1
Token-Level Diagnosis of Sycophancy in LLMs with Attribution-Guided Steering
Jul 31, 2026Hieu Nguyen, Mahammed Kamruzzaman, Anshuman Chhabra +1LLM Sycophancy LLM Interpretability
Bellini College of AI, Cybersecurity and Computing University of South Florida
A Cross-Architecture Audit of Direction-Based Inference-Time Defences in Vision-Language Models
Jul 30, 2026Xiangyu Yin, Tora Bodin, Rohan Menon +1VLM Robustness Adversarial Attacks on VLMs
Chalmers University of Technology, Gothenburg, Sweden · Carl von Ossietzky University of Oldenburg, Oldenburg, Germany
Policy Gradient Steering: Interventions from Behavioral Objectives
Jul 30, 2026Yoann Poupart, Aurélie Beynier, Nicolas MaudetPolicy Gradient Activation Steering
LIP6, Sorbonne University Paris, France
Do Unified Multimodal Models Think in One Space? A Lens Through Cross-Branch Steering
Jul 29, 2026Yu Wang, Sharon LiUnified Multimodal Models Cross-Modal Alignment
Department of Computer Sciences, University of Wisconsin-Madison
Misalignment Has a Personality: A Big Five Account of Emergent Misalignment
Jul 29, 2026Hasibur Rahman, Smit DesaiLLM Alignment Personality Modeling in Language Models
Northeastern University Boston, MA, US
Minimizing Targeted Activations: Input-Only Suppression of Evaluation-Awareness Latents in Large Language Models
Jul 28, 2026Deepanshu Mody, Samarth Agarwal, Utkarsh Mittal +1Language Model Safety Evaluation LLM Interpretability
Center for Data Science, New York University
Forecasting Side Effects of Activation Steering
Jul 28, 2026Chong Yong Ong, Alson Wei Jie Sim, Peixin Zhang +1Language Model Steering LLM Auditing
Singapore Management University
Where Steering Signals Come From: Activation Source Selection in Activation Steering
Jul 28, 2026Jiaran Ye, Lingxu Ran, Zijun Yao +5Language Model Steering Activation Steering
1MOE Key Laboratory of Computational Linguistics, Peking University · Department of Computer Science and Technology, Tsinghua University · 6YiXin-AILab, YIXIN, Beijing, China +3
OPIUM: Mitigating Steering Externalities and Over-Refusal via Dual Objective Latent Optimization
Jul 22, 2026Kavin Aravindan, Arihant Rastogi, Krishak Aneja +4LLM Safety Alignment Over-Refusal
IIIT Hyderabad, India.
Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering
Jul 20, 2026Sheldon Yu, Tong Yu, Xunyi Jiang +6Overthinking in Language Models Causal Interventions in Language Models
UC San Diego · Adobe Research · University of New South Wales
First-Order Predictable but Pairwise Fragile: Local Task Adaptation in Trained Transformers
Jul 18, 2026Irina Piontkovskaia, Sergey NikolenkoFine-Tuning Neural Network Optimization
1DAIMLD, Moscow, Russia · 2St. Petersburg Department of the Steklov Institute of Mathematics, St. Petersburg, Russia · 3St. Petersburg State University
Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal Control
Jul 16, 2026Jihoon Hong, Julian Skifstad, Qiyue Dai +2World Models for Robotics Action-Conditioned World Models
Georgia Institute of Technology
Graded Entity-Familiarity Readouts in Language Models: Polish Adaptation, Cross-Language Robustness, and Refusal Steering
Jul 15, 2026Grzegorz BrzezinkaMultilingual Language Model Evaluation Language Model Introspection
Prosit AS
Encoder-Side Neuron Identification and Amplification for Acoustic Perception in Large Audio-Language Models
Jul 13, 2026Yu-Han Huang, Chih-Kai Yang, Ke-Han Lu +2Audio Understanding Audio-Language Models
National Taiwan University, Taiwan · Graduate Institute of Communication Engineering, National Taiwan University, Taiwan · NTU Artificial Intelligence Center of Research Excellence (NTU AI-CoRE), Taiwan
Conditional Optimal Bridge for Riemannian Activation Steering
Jul 12, 2026Seyed Arshan Dalili, Ajay Narayanan Sridhar, Vijaykrishnan Narayanan +1Activation Steering
The Pennsylvania State University
SPARK: Susceptibility-Guided Profiling and Steering of Latent Reasoning States in Large Language Models
Jul 11, 2026Dongxu Zhang, Yiding Sun, Zihao Guo +5LLM Interpretability Activation Steering
Xi’an Jiaotong University · Peking University · Tencent
Present but Rescaled: Chat-to-Agent Transfer of Additive Activation Steering
Jul 10, 2026Lucas PintoTool-Using Agents LLM Agent Safety
Independent Researcher
LLMs Silently Correct African American English: Auditing and Mitigating Dialect Bias via Activation Steering
Jul 7, 2026Huan Wu, Ali Emami, Muhammad Furquan Hassan +5Language Model Steering Social Bias in Language Models
York University · Vector Institute · Connected Minds +6
Controlling Tool Use with Heading-Specific Activation Steering
Jul 7, 2026Yuqi Chen, Vincent Siu, Yang Liu +2Language Model Steering LLM Tool Use
Department of Computer Science and Engineering, UC Santa Cruz · Department of Computer Science, UC Berkeley
A Coin Flip Per Token: Bernoulli Sparse Steering of Large Language Models
Jul 6, 2026Nima Eshraghi, Lovedeep Gondara, Yuqing Huang +5Language Model Steering Sparse Autoencoders
The Vanguard Group, Inc. · University of Toronto
Unsupervised Features Mining via Activation Geometry
Jul 5, 2026Amit LeVi, Elad David, Max FominLLM Interpretability Unsupervised Representation Learning
Zenity Tel Aviv, Israel
Can Dialects Be Steered Like Languages? Sparse Neurons and Distributed Directions in Arabic LLMs
Jul 4, 2026Kareem Elozeiri, Mervat Abassy, Omar Kallas +4Arabic NLP LLM Interpretability
1Mohamed bin Zayed University of Artificial Intelligence · 2Qatar Computing Research Institute, Hamad Bin Khalifa University · 3Tohoku University +1
FairFlow: Demystifying and Mitigating Stereotype Bias in Text-to-Image Diffusion Transformers
Jul 3, 2026Chen Chen, Yuanmin Huang, Zhenfei Zhang +5Diffusion Model Alignment Diffusion Transformer
Fudan University Shanghai, China · East China University of Science and Technology Shanghai, China
Harnessing the Latent Space: From Steering Vectors to Model Calibrators for Control and Trust
Jun 30, 2026Nishant SubramaniLanguage Model Steering LLM Interpretability
Carnegie Mellon University, Language Technologies Institute
Harnessing Textual Refusal Directions for Multimodal Safety
Jun 30, 2026Moreno D'Incà, Nicu Sebe, Massimiliano ManciniLanguage Model Safety Evaluation Multimodal Large Language Models
University of Trento
Internal-State Probes Read the Situation, Not the Action: Three Negative Results for Pre-Action Misalignment Monitoring
Jun 29, 2026Max Fomin, Elad David, Amit LeViLanguage Model Safety Evaluation AI Agent Monitoring
Zenity
Neural Procedural Memory: Empowering LLM Agents with Implicit Activation Steering
Jun 29, 2026Chengfeng Zhao, Yuqiao Tan, Shizhu He +3LLM Agent Memory LLM Agent Skill Learning
Institute of Automation, CAS · University of Chinese Academy of Sciences · Beijing Academy of Artificial Intelligence
Closing the Activation-Cone Blind Spot: Response-Time Probing and Unified Defense
Jun 28, 2026Subhadip MitraAdversarial Attacks on LLMs LLM Safety
Independent Researcher

Jul 31, 2026J/K move · Enter open · S save

A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use

Yutong Ke, Ming Yin, Chongwen Zhao, Kaizhu Huang

LLM Interpretability LLM Tool Use Activation Steering LLM Agent Reliability

Open paper AskPDF arXiv

Agentic LLMs exhibit three consequential tool-use failures: invalid arguments (validity), unnecessary calls (over-calling), and omitted calls when tools are needed (missing). We find that a small, failure-specific set of MLP neurons could distinguish such failures with linearly separable decision boundaries. Building on this observation, we introduce PRISMS (Probing Representations In Support of Monitoring and Steering), a closed-loop framework that shares a failure-specific neuron basis between sparse detection and activation steering. PRISMS selects contribution-critical MLP neurons and fits an L1-regularized detector on their activations. Across six models from the Qwen3, Llama, and Gemma families, over-calling and missing are detected at the pre-generation prompt boundary with ROC-AUC 0.90-1.00, while validity is detected from the generated tool-call span with ROC-AUC 0.86-0.90. These results are achieved with highly sparse readouts: only 1-2 MLP neurons for missing, 2-16 for over-calling, and approximately 128 for validity. These sparse detectors match or outperform dense residual-stream baselines using 23-627 times fewer features. The shared neuron basis also supports bidirectional control over tool-calling behavior, suppressing unnecessary calls and eliciting omitted ones. PRISMS therefore gates intervention on predicted failure risk to mitigate the collateral effects of unconditional steering. Across all six models, PRISMS reduces pooled over-calling rate by 80% (from 0.131 to 0.026) while increasing tool-required accuracy by 14.2 percentage points (from 0.689 to 0.831). PRISMS thus provides lightweight failure detection and selective intervention across model families.