Harms

Follow this topic Ask about this topic

Momentum

8 papers in the last four weeks, up 100% on the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 62

All topics

Tool Mediation Alters Refusal Mechanisms in Large Language Models
Sep 28, 2026Abel Rodríguez, Giuseppe Garofalo, Lieven Desmet +1Large Language Models Fail Refusals
DistriNet, KU Leuven
From Attack Success to Attack Severity: Counterfactual Memory Attacks on LLM Agents
Sep 28, 2026Mingxi Zou, Langzhang Liang, Zhuo Wang +3Large Language Model Agents Harms
Shanghai Academy of AI for Science (SAIS) · Fudan University · Monash University
Path-specific harm decomposition: A partial identification framework
Sep 24, 2026Ruizi Yan, Dennis Frauen, Maresa Schröder +1Causal Mediation Analysis Harms
Department of Statistics, University of Oxford · LMU Munich · Munich Center for Machine Learning (MCML)
When Post-Processing Fairness Constraints Help and When They Harm: Evidence from Eight Cross-Domain Evaluations
Sep 22, 2026Nithin Raghava Ramachandra NarlaFairness Audits Algorithmic Fairness
Independent Researcher, Dallas-Fort Worth, Texas, United States
ClashBench: Conflicts Leading Agents to Seize and Harm
Sep 17, 2026Yuejin Xie, Yu Li, Dadi Guo +6Runtime Safety Filtering Preemption
Tsinghua University · Shanghai AI Lab · Fudan University +2
Safety Beyond the Interface: Detecting Harm via Latent States in Large Language Models
Sep 16, 2026Alizishaan Khatri, Chiquita Prabhu, Omkar NeogiLarge Language Model Safety Guardrail
Wrynx Inc. · Independent Researcher
Characterizing Bluesky Content Moderation Service: From Automation of Service to Landscape of Harms
Sep 12, 2026Pushpdeep Singh, Sayeh Jarollahi, Ayan Majumdar +5Content Moderation Model Auditing
Max Planck Institute for Software Systems (MPI-SWS), Germany · Saarland University, Germany · Indian Institute of Technology Kharagpur, India
Signed Rescue Routing: Harm-Aware Cascades for Efficient LLM Inference
Sep 7, 2026Zheyuan Wang, Siyu Li, Peiqiao Song +3Large Language Model Routing LLM Inference Optimization
School of Artificial Intelligence, Beijing Normal University · School of Information Science and Engineering, Chongqing Jiaotong University · School of Computer Science and Technology, Dalian University of Technology +1
HarmoCore: Functional Latent Diffusion for Sparse Reconstruction of Oscillatory Wave Fields
Sep 1, 2026Lihao Chen, Xinyu Zhang, Panqi Chen +4Complex Wavefield Signal Reconstruction
College of Information Science and Electronic Engineering, Zhejiang University
The Safety Relay in Roleplay Jailbreaks: A Component-Resolved Causal Analysis of Harm Recognition and Refusal
Aug 31, 2026Md Mokarram Chowdhury, Ernie Chang, Yang LiJailbreaks Jailbreak Attacks
Department of Computer Science, Iowa State University, United States · Meta, United States
AI and Consumer Rights in India Working Paper
Aug 13, 2026Omir Kumar, Sriya Sridhar, Vibhav Mithal +1Artificial Intelligence Systems Hindi
Centre for Responsible AI IIT Madras Chennai, India
Refusing Everything Looks Safe: Restoring the Benign Arm to Encoded-Prompt Evaluation
Aug 12, 2026Haoyu Zhang, Haowen Xu, Xiao Luo +7Large Language Model Safety Benign
Northeastern University
HarMoE: Multi-Source Chest Radiograph Pretraining with Dataset-Disentangled Experts
Aug 3, 2026Haozhe Luo, Ziyu Zhou, Shelley Zixin Shu +1Chest X-Ray Medical Vision-Language Models
ARTORG Center for Biomedical Engineering Research, University of Bern · Kaiko.AI, Zurich, Switzerland · Shanghai Jiao Tong University, Shanghai, China
(Im)Paired Programming: Coding Agents Improve Productivity but Harm Understanding
Jul 29, 2026Nishant Balepur, Connor Baumler, Valerie Chen +3Coding Agents Artificial Intelligence Agents
University of Maryland · New York University · Carnegie Mellon University
Psychological Influences of Conversational AI: Research and Design Directions for Reducing Harm and Promoting Well-Being
Jul 27, 2026Jina Suh, Mihaela Vorvoreanu, Forough Poursabzi-Sangdeh +5Well-Being Conversational Artificial Intelligence
Microsoft, USA · University of Washington · City University of New York, USA +1
Harm is not Universal: Community-Specific Toxicity Detection is Urgently Needed
Jul 27, 2026Xinnuo Xu, Anja Thieme, Daniela Massiceti +6Toxicity Adversarial Prompts
Microsoft Research Cambridge, UK · Microsoft Paris, France
When Should Active RAG Retrieve? A Budget-Aware Evaluation of Utility, Calibration, and Cost
Jul 27, 2026Pin Qian, Su Wang, Chong Peng +5Retrieval-Augmented Generation Pipelines Token Budget Allocation
Carnegie Mellon University Pittsburgh, PA, USA · University of Glasgow Glasgow, United Kingdom · Independent Researcher Seattle, WA, USA +1
Sound Probabilistic Safety Bounds for Large Language Models
Jul 22, 2026Mahdi Nazeri, Anne-Kathrin Schmuck, Sadegh Soudjani +1Large Language Model Safety Probabilistic Safety
Department of Computer Science, University of Oxford, UK · Max Planck Institute for Software Systems, Germany · School of Computer Science, University of Birmingham, UK
HarmoHOI: Harmonizing Appearance and 3D Motion for Multi-view Hand-Object Interaction Synthesis
Jul 19, 2026Lingwei Dang, Juntong Li, Zonghan Li +5Hand-Object Interaction Detection 3D Motion
South China University of Technology, Guangzhou, China · Beijing Normal University, Beijing, China · Tsinghua University, Beijing, China +1
AMT-X: Phase-Structured Multi-Turn Red-Teaming with Checklist-Gated Evaluation
Jul 13, 2026Yi Ting Shen, Kentaroh Toyoda, Alex LeungRed-Teaming Attack-Success Rate
Vulcan Research, AIFT, Singapore
ANCHOR: Automated Alignment Auditing for CLI Agents on Real-World Harm
Jul 11, 2026Kefan Song, Yanjun QiAgentic Deployments Malicious Agents
University of Virginia, Charlottesville, VA, USA
It Lied to a Doctor to Buy Poison Ingredients: Quantifying Real-World Misuse of Phone-use Agents
Jun 26, 2026Yiming Sun, Chen Chen, Zifan Zhou +1Computer-Use Agents Artificial Intelligence Agents
Fudan University, JADE (Whitzard AI) Team
HarmVideoBench: Benchmarking Harmful Video Understanding in Large Multimodal Models
Jun 25, 2026Jiajun Wu, Haoyu Kang, Yining Sun +13Long-Video Benchmarks Large Multimodal Models
Central South University · Tsinghua University · South China Normal University +10
One Year Later...The Harms Persist, But So Do We!
Jun 22, 2026Annika Marie Schoene, Cansu Canca, Gautham Vijay Kumar +1Mental Health Large Language Model Safety
Northeastern University Boston, MA, USA · Florida International University Miami, FL, USA
On Defining Erasure Harms for NLP
Jun 14, 2026Yu Lu Liu, Arnav Goel, Jackie Chi Kit Cheung +3Concept Erasure Harms
1Johns Hopkins University · 2Carnegie Mellon University · 3Mila – Québec Artificial Intelligence Institute +2
HarmoView: Harmonizing Multi-View Constraints for Identity-Consistent Video Generation
Jun 9, 2026Cong Wang, Zhentao Yu, Hongmei Wang +7Long Video Generation Multi-View
Tencent Hunyuan
The Distributed Detectability Band Against Marginal-Preserving Attacks
Jun 9, 2026Zhang Qinqin, Gao YuzeSabotage Distributed Denial-Of-Service
No-Harm Physics-Informed Inverse Learning with Residual-Calibrated Uncertainty
Jun 5, 2026Ronald KatendeCalibrated Uncertainty Reconstruction Error
Department of Mathematics Kabale University Kikungiri Hill, Katuna Road, Kabale, Uganda
Korean Culture into LLM Alignment: Toward Cultural Coherence
Jun 5, 2026MinJae Jung, Minwoo KimLarge Language Model Alignment Large Language Model Safety
DATUMO Inc.
Investigating and Alleviating Harm Amplification in LLM Interactions
Jun 1, 2026Ruohao Guo, Wei Xu, Alan RitterLarge Language Model Safety Harms
Georgia Institute of Technology
LLM Judges Inconsistently Disagree Across Safety Criteria and Harm Categories
May 29, 2026Krishnapriya Vishnubhotla, Sowmya Vajjala, Akriti Vij +1Large Language Model Judges Judges
National Research Council, Canada · IMDA, Singapore
MuPHI: Learning Implicit Multimodal Harm Reasoning via Semantically Grounded Reward Optimization
May 28, 2026Anisha Saha, Varsha Suresh, Teodora Kamova +3Multimodal Reasoning Cross-Modal
Max Planck Institute for Informatics, Saarland Informatics Campus · Saarland University · The University of Edinburgh +1
Harmless Yet Harmful: Neutral Prompting Attacks for Stealthy Hallucination Steering in Agent Skills
May 28, 2026Chia-Yi Hsu, Chia-Mu Yu, Chun-Ying Huang +1Malicious Agents Coding Agents
Department of Computer Science National Yang Ming Chiao Tung University · Department of Electronics and Electrical Engineering National Yang Ming Chiao Tung University · School of Computing Institute of Science Tokyo
HarmoVid: Relightful Video Portrait Harmonization
May 27, 2026Jun Myeong Choi, Jae Shin Yoon, Luchao Qi +2Video Restoration Video Diffusion Models
University of North Carolina at Chapel Hill · Adobe Research
PAST2HARM: A Simple Adaptive Past Tense Attack for Jailbreaking Multimodal AI
May 26, 2026Snehasis MukhopadhyayAdversarial Prompts Harms
Indian Institute of Information Technology, Kalyani
HARP: Measuring Harm Amplification in Multi-Agent LLM Systems
May 26, 2026Md Hafizur Rahman, Zafaryab Haider, Tanzim Mahfuz +1Multi-Agent Large Language Model Systems Harms
Electrical and Computer Engineering University of Maine, ME, USA
First, do no harm: Breaking suicidogenic echo chambers in media recommendation
May 24, 2026Alberto Díaz-Álvarez, Raúl Lara-Cabrera, Fernando Ortega-Requena +1Suicide Risk Recommendation
E.T.S.I. Sistemas Informáticos (Universidad Politécnica de Madrid), C. de Alan Turing, s/n, 28031 Madrid, Spain
Counterfactually Safe Reinforcement Learning
May 24, 2026Jingyi Li, Peng Wu, Chengchun ShiCounterfactual Learning Optimality
Department of Statistics and Data Science, National University of Singapore · School of Mathematics and Statistics, Beijing Technology and Business University · Department of Statistics, London School of Economics and Political Science
Do No Harm? Hallucination and Actor-Level Abuse in Web-Deployed Medical Large Language Models
May 20, 2026Sunday Oyinlola Ogundoyin, Muhammad Ikram, Rahat MasoodHallucination Detection Harms
Macquarie University, Sydney, Australia · The University of New South Wales, Sydney, Australia
DisaBench: A Participatory Evaluation Framework for Disability Harms in Language Models
May 12, 2026Eugenia Kim, Ioana Tanase, Christina MallonSafety Benchmarks Disability
Microsoft
HarmoWAM: Harmonizing Generalizable and Precise Manipulation via Adaptive World Action Models
May 11, 2026Qiuxuan Feng, Jiale Yu, Jiaming Liu +8Efficient World-Action Model World Models
State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University · The Chinese University of Hong Kong · Simplexity Robotics
Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms
May 8, 2026Linh Le, David Williams-King, Mohamed Amine Merzouk +2Adversarial Training Harms
McGill University · Mila – Quebec AI Institute · ERA +1
The Moltbook Files: A Harmless Slopocalypse or Humanity's Last Experiment
May 8, 2026William Brach, Federico Torrielli, Stine Lyngsø Beltoft +3Large Language Model Safety Openclaw
University of Turin Dept. of Computer Science Torino, Italy · University of Southern Denmark Odense, Denmark
Gaming the Metric, Not the Harm: Certifying Safety Audits against Strategic Platform Manipulation
May 7, 2026Florian A. D. Burnat, Brittany I. DavidsonSafety Benchmarks Security Evaluation
University of Bath
One Turn Too Late: Learning When to Intervene Against Multi-Turn Malicious Intent
May 7, 2026Xinjie Shen, Rongzhe Wei, Peizhi Niu +6Attacker Malicious Agents
Georgia Institute of Technology · University of Illinois Urbana-Champaign · UCSD +3
HARMES: A Multi-Modal Dataset for Wearable Human Activity Recognition with Motion, Environmental Sensing and Sound
May 4, 2026Robin Burchard, Pascal-André Brückner, Marius Bock +2Human Activity Recognition Multi-Modal Sensing
University of Siegen, Siegen, Germany · University of Bonn and & Lamarr Institute for Machine Learning and Artificial Intelligence, Bonn, Germany
PrismAgent: Illuminating Harm in Memes via a Zero-Shot Interpretable Multi-Agent Framework
May 1, 2026Zihan Ding, Ziyuan Yang, Yi ZhangMemes Hate Speech Detection
Sichuan University, Chengdu, Sichuan, China · Nanyang Technological University, Singapore
From Prompt Risk to Response Risk: Paired Analysis of Safety Behavior of Large Language Models
Apr 28, 2026Mengya Hu, Qiong Wei, Sandeep AtluriLarge Language Model Safety Harms
Responsible AI, Microsoft
Representational Harms in LLM-Generated Narratives Against Global Majority Nationalities
Apr 24, 2026Ilana Nguyen, Harini Suresh, Thema Monroe-White +1Large Language Model Bias Narratives
Brown University, USA · Brown University, Providence, Rhode Island, USA · George Mason University, Fairfax, Virginia, USA +1
When Safety Fails Before the Answer: Benchmarking Harmful Behavior Detection in Reasoning Chains
Apr 21, 2026Ishita Kakkar, Enze Zhang, Rheeya Uppaal +1Reasoning Traces Reasoning Chain
Department of Computer Sciences, University of Wisconsin-Madison
Human-Guided Harm Recovery for Computer Use Agents
Apr 20, 2026Christy Li, Sky CH-Wang, Andi Peng +1Computer-Use Agents Harms
MIT CSAIL · Abridge · humans&
Informing AI Policy Assessment using Large-Scale Simulation of Interventions
Apr 20, 2026Julia Barnett, Kimon Kieslich, Natali Helberger +1Eu Artificial Intelligence Act Artificial Intelligence Governance
Northwestern University, USA · University of Amsterdam, The Netherlands · University of Hohenheim, Germany
Owner-Harm: A Missing Threat Model for AI Agent Safety
Apr 20, 2026Dongcheng Zhang, Yiqing JiangHarms
BlueFocus Communication Group · Tongji University
First, Do No Harm (With LLMs): Mitigating Racial Bias via Agentic Workflows
Apr 20, 2026Sihao Xing, Zaur GoulievLarge Language Model Bias Diagnosis
School of Enterprise Computing and Digital Transformation Technological University Dublin · School of Information and Communication Studies University College Dublin
RoTRAG: Rule of Thumb Reasoning for Conversation Harm Detection with Retrieval-Augmented Generation
Apr 19, 2026Juhyeon Lee, Wonduk Seo, Junseo Koh +3Harms Dialogue Benchmarks
Peking University · Enhans; Peking University · Enhans +1
DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training
Apr 18, 2026Ziwen Pan, Zihan Liang, Jad Kabbara +1Large Language Model Safety Harms
Emory University · MIT
Large Language Models Generate Harmful Responses Using a Distinct Mechanism, Shared Across Harm Types
Apr 10, 2026Hadas Orgad, Boyi Wei, Kaden Zheng +4Large Language Model Safety Harmful Content
Kempner Institute, Harvard University · Princeton University · Harvard University +2
A Marketplace for AI-Generated Adult Content and Deepfakes
Jan 14, 2026Shalmoli Ghosh, Matthew R. DeVerna, Filippo MenczerDeepfake Detection Harms
Indiana University Bloomington · Stanford University
First, do NOHARM: a medical safety benchmark and randomized study of physician and AI teaming on clinical consultations
Dec 1, 2025David Wu, Fateme Nateghi Haredasht, Saloni Kumar Maharaj +54Harms Ai-Based
ECHO: A Participatory Framework for Bias-Anchored AI Harm Anticipation
Nov 27, 2025Nicoleta Tantalaki, Sophia Vei, Athena VakaliEthics Harms
Aristotle University, Thessaloniki, Greece

Sep 28, 2026, cs.AIJ/K move · Enter open · S save

Tool Mediation Alters Refusal Mechanisms in Large Language Models

Abel Rodríguez, Giuseppe Garofalo, Lieven Desmet, Vera Rimmer

DistriNet, KU Leuven

Large Language Models Fail Refusals Mediation Harms

Open paper AskPDF arXiv

Large language models (LLMs) are increasingly deployed with access to external tools, yet harmful tool-mediated interactions are less likely to be refused when compared to regular conversational ones. As this change in refusal behavior remains underexplored, we investigate its underlying mechanisms across a diverse set of open-weight language models. We find that information about the harmfulness of a request remains strongly encoded in the model's representations and transfers across conversational and tool-mediated inputs. Evidence from representation geometry and neuron-level analysis further indicates that the two interaction modes systematically distribute harm-related computation differently. Crucially, while conversational inputs can be refused at relatively low levels of perceived harmfulness, tool-mediated inputs remain permissive until harmfulness crosses a substantially higher effective refusal threshold. Moreover, tool-mediated refusal is also more brittle: progressively weakening the refusal computation disrupts tool-mediated refusal at lower intervention strengths than conversational refusal, even when benign capabilities remain intact. Together, our findings indicate that tool mediation does not simply reduce the internal perception of harm, but instead impacts its conversion into refusal. Overall, this suggests tool-mediated environments may intrinsically reduce robustness of models to harmful requests, and that conventional safety evaluations may not fully transfer to LLM agents.

Momentum

Related topics

Latest papers 62