LLM Safety

LLM: Large Language Model

LLM safety studies how models can avoid harmful assistance, manipulation, privacy violations, and unreliable advice while still helping with legitimate requests. Approaches include safety training and data filtering, inference-time guardrails and steering, output monitoring, and mechanistic methods that probe or edit internal representations. Benchmarks increasingly test jailbreaks across languages and modalities, long conversations, tool use, and high-stakes settings such as health and cybersecurity. A central challenge is maintaining robustness against adaptive attacks and changing contexts without causing over-refusal or sacrificing utility.

Latest papers

442 papers68 papers in the last four weeks, up 70% on the four weeks before. 0.4% of all new papers.

One Vector or Many? Clustered Optimal-Transport Steering under Quantization
Oct 29, 2026Illia MakarovActivation Steering LLM Safety
Faculty of Informatics, National University of Kyiv-Mohyla Academy
The Role of Self-Concept In Language Models
Oct 29, 2026Yijie Hao, Joyce C. HoLLM Safety
Department of Computer Science, Emory University, Atlanta, GA, USA
CANS: Capability-Preserving Neuron Steering for Large Language Models
Oct 26, 2026Yucheng Wu, Chengyi Xie, Chenpeng Wang +1Language Model Steering Safety-Utility Trade-Off in Language Models
State Key Laboratory of Multimedia Information Processing, Peking University · School of Computer Science, Peking University · YiXin-AILab, YIXIN, Beijing, China
Trustworthy Recursive Self-Improvement in the Era of Large Models: Foundation, Risks, Evaluation and Open Challenges
Oct 26, 2026Jiayue Pu, Ang Gao, Jinyi Bai +9Recursive Self-Improvement Language Model Self-Improvement
Gaoling School of Artificial Intelligence, Renmin University of China · University of Chinese Academy of Sciences
C-SafeGen: Certified Safe LLM Generation with Claim-Based Streaming Guardrails
Oct 26, 2026Mintong Kang, Zhaorun Chen, Bo LiLLM Safety LLM Guardrails
UIUC · UChicago · UIUC & UChicago
When Personas Become Harmful: Guiding LLMs with Global Citizenship Values Alignment
Oct 24, 2026Siqi Shen, Lajanugen Logeswaran, Moontae Lee +2LLM Safety LLM Safety Alignment
University of Michigan · LG AI Research · University of Illinois at Chicago
SEDATE: Do Large Language Models Share an Aggression Circuit Across Languages?
Oct 24, 2026Sagnik Basu, Sayan Das, Somnath Banerjee +1LLM Safety Language Model Editing
Indian Institute of Technology Kharagpur · Singapore Institute of Technology
REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories
Oct 23, 2026Hanwen Zhang, Enyi Jiang, Ruike Zhu +2LLM Safety Causal Interventions in Language Models
Independent Researcher · Stanford University · University of Illinois Urbana-Champaign
Faithful Latent Harm Monitors Work No Better Than the Models Own Behaviour at a Matched False Alarm Rate
Oct 22, 2026Abhishek Mishra, Sagnik Chatterjee, Suyash ManiyarLLM Safety Monitoring Language Model Safety Evaluation
University of Massachusetts Amherst
From Clinical Judgment to Checkable Rules: Clinician-Led Guardrails for Mental-Health Chatbots
Oct 22, 2026Chenhan Lyu, Kit Wislocki, Sarah Elizabeth Stoeckl +2LLM Guardrails LLM Safety
University of California, Irvine, USA
Probing and Steering for Safety-Relevant Behaviours in Large Language Models
Oct 21, 2026Nimun Bajwa, Massimo PoesioActivation Steering Language Model Probing
Queen Mary University of London
Fooling LLM Safety Guards into Judging the Wrong Text
Oct 20, 2026Utkarsh KanwatLLM Guardrails Adversarial Attacks on LLMs
Independent Researcher
Can Native Safety Travel? Reusing Safety Pretraining Across Models via On-Policy Distillation
Oct 20, 2026Bingqi Shang, Yang Zhang, Yiwei Chen +4LLM Safety
Michigan State University · MIT-IBM Watson AI Lab · University of California, Santa Barbara +1
Global Security Blind Spots: Translation as a Safety Bypass in Low-Resource Languages
Oct 19, 2026Jalil Alva, David Harper, Abigail E. Thornton +8LLM Safety Language Model Safety Evaluation
Welo Data
Safety Mid-Training: Internalizing LLM Safety as a Foundational Capability
Oct 19, 2026Zhengyue Zhao, Liwei Jiang, Yejin Choi +1Safety-Utility Trade-Off in Language Models LLM Safety Alignment
Johns Hopkins University · University of Washington · Stanford University
Reasoning-Aware Multicalibrated Reliability Predictors for Language Models
Oct 19, 2026Ekaterina Fedorova, Alessandro Morosini, Chara PodimataLanguage Model Calibration LLM Reliability
Massachusetts Institute of Technology.
AfriGuard: Safety and Robust Refusal For African Languages
Oct 19, 2026Israel Abebe Azime, Jesujoba O. Alabi, Iffat Maab +20LLM Safety Language Model Safety Evaluation
Saarland University, Saarland Informatic Campus, Germany · Masakhane NLP · National Institute of Informatics, Japan +13
Safety Kernel: Mechanistic Alignment for Cross-Lingual Defense in LLMs
Oct 17, 2026Chengyu Xie, Fei Shen, Shuyi Miao +5LLM Safety Language Model Robustness
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, China · Nanjing University of Science and Technology, Nanjing, China · School of Artificial Intelligence, Beihang University, Beijing, China +3
Refusal Is Not Robustness: Measuring Harmful Reachability in Autoregressive Language Models
Oct 15, 2026Parv Kapoor, Teresa Fortes, Claudia Mamede +3Language Model Safety Evaluation LLM Safety
Carnegie Mellon University, Pittsburgh, PA, USA. · General Robotics · University of Porto, Porto, Portugal.
Risk and Refusal: Mechanisms of LLM Responses to Suicidal Ideation
Oct 15, 2026Drishti Goel, Soorya Ram Shimgekar, Jeongah Lee +4LLM Safety LLM Refusal Behavior
University of Illinois Urbana-Champaign, USA · University of Massachusetts Amherst, USA · OSF HealthCare
What Social Media Teaches Us About Accumulative AI Risks
Oct 15, 2026Dawei Li, Bohan Jiang, Huan LiuAI Safety LLM Safety
School of Computing and Augmented Intelligence, Arizona State University
When Reasoning Is Not Enough: Proactive Inquiry for Cross-Modal Safety in Multimodal Large Language Models
Oct 15, 2026Wei Cai, Jian Zhao, Yuchen Yuan +1Multimodal Multi-Turn Safety Alignment VLM Safety
Institute of Artificial Intelligence (TeleAI), China Telecom, P. R. China. · Peking University. · School of Artificial Intelligence, OPtics and ElectroNics (iOPEN), Northwestern Polytechnical University, Xi’an 710072, P. R. China.
Recoverability-Controlled Safety Miscalibration in Large Language Models
Oct 15, 2026Yuwei Wang, Xingwang Liang, Yu Xie +2Language Model Safety Evaluation LLM Safety
School of Automation and Software Engineering, Shanxi University · School of Computer and Information Technology, Shanxi University
Resisting Harmful Instructions through Training-Free Weight Editing
Oct 15, 2026Ali Ebrahimpour-Boroojeny, Xiaocong Yang, Samuel Schapiro +4LLM Safety LLM Safety Alignment
PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models
Oct 14, 2026Giovana Kerche Bonás, Thales Sales Almeida, Marcelo da Silva Reis +2LLM Safety Causal Interventions in Language Models
RECOD.ai Lab, Institute of Computing, Universidade Estadual de Campinas · Maritaca AI
Safe-FCL: Continual Risk Awareness and Fine-Grained Safety Aggregation for Federated Continual Learning of LLMs
Oct 14, 2026Shuaijie Ma, Yongkang Hu, Yu Cheng +4Federated Continual Learning Federated Learning
East China Normal University · Shanghai Innovation Institute
DANGERLENS: Probing Dangerous Representations Hidden by Safety Alignment
Oct 14, 2026Tianzhuo Yang, Shaoheng Yan, Lin Lu +5LLM Safety LLM Safety Alignment
Institute for Artificial Intelligence, Peking University · JIUTIAN Research, China Mobile
Genesis Steering: Better Activation Steering Through Training Data
Oct 13, 2026Tora Bodin, Eric Aubinais, Ilaria Torre +1Language Model Steering Activation Steering
Chalmers University of Technology, Gothenburg, Sweden · Carl von Ossietzky University of Oldenburg, Oldenburg, Germany
How Safe Is ChatGPT Health Advice? Evidence from Real User Conversations in Four Global South Countries
Oct 13, 2026Shreyasi Roy Chowdhury, Kiran GarimellaLLM Safety Language Model Safety Evaluation
Independent Researcher · Rutgers University
Efficient Safety Guardrails through Latent Dual-Hypothesis Reasoning with On-Demand Trace Reconstruction
Oct 13, 2026Md Asiful Islam, Fahmida Alam, Mihai SurdeanuLLM Guardrails LLM Safety
Department of Computer Science University of Arizona, USA

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

One Vector or Many? Clustered Optimal-Transport Steering under Quantization

Illia Makarov

Faculty of Informatics, National University of Kyiv-Mohyla Academy

Activation Steering LLM Safety Optimal Transport Causal Interventions in Language Models

Open paper AskPDF arXiv

Activation steering changes what a language model generates by editing its hidden states at inference time. Its most common form adds one vector, the difference of class means (DiM), to every hidden state. CHaRS (concept heterogeneity-aware representation steering) uses optimal transport instead: an entropic plan pairs the clusters of the two classes, and each hidden state moves along a mixture of the cluster displacements, weighted toward its nearest clusters. We study this family of clustered optimal-transport maps for toxicity mitigation in Gemma-2-2B and Granite-3.1-2B, and ask what the plan adds to DiM and whether that survives INT8 and NF4 quantization. We prove an identity: whatever the plan, the displacements averaged with the cluster sizes as weights equal DiM, so a clustered map can differ from DiM only for tokens whose routing weights depart from the cluster sizes. In the CHaRS protocol they stay close, and CHaRS is statistically equivalent to Mean-AcT, which adds DiM at every layer, in all eight pre-specified equivalence tests. A variant fitted on individual tokens, which routes each token mainly to its nearest clusters, leaves 37-39% less toxicity than DiM on Gemma and 51-61% less on Granite at steering strength 3 (toxic-bert), where DiM already removes 76-85% of it. Two other toxicity classifiers confirm the gain on Granite, but on Gemma only one does. At equal strength the variant has somewhat higher perplexity, but at matched toxicity it is slightly more fluent than DiM. Yet a nearly constant map from the same clusters also leaves about 30% less toxicity, so the role of treating tokens differently remains open. Under INT8 and NF4 the variant still leaves less toxicity than DiM, and transferring it from full precision to the NF4 model instead of refitting it adds no significant toxicity after correction for multiple tests.