LLM Safety Benchmarks

LLM: Large Language Model

This field measures how reliably language models recognize and handle risks such as harmful requests, jailbreaks, bias, privacy violations, and unsafe tool use. Benchmarks use curated and expert-annotated scenarios, adversarial red-teaming, multi-turn and agent simulations, and increasingly multilingual, multimodal, and domain-specific tests. Researchers also compare refusal and moderation methods, use repeated sampling and statistical techniques to estimate reliability, and assess both harmful compliance and over-refusal. A central challenge is making results comparable and trustworthy while ensuring tests reflect changing, context-dependent behavior in real deployments.

Latest papers

284 papers44 papers in the last four weeks, up 7% on the four weeks before. 0.3% of all new papers.

Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs
Apr 8, 2026Yue Huang, Haomin Zhuang, Jiayi Ye +6LLM Guardrails LLM Safety Alignment
University of Notre Dame · University of California, Los Angeles · MIT-IBM Watson AI Lab +1
LURE: Live-Usage Replay Evaluations for Reducing Evaluation Awareness
Apr 8, 2026Igor Ivanov, David Demitri AfricaLLM Evaluation LLM Safety Evaluation
Meridian Cambridge
GrandGuard: Taxonomy, Benchmark, and Safeguards for Elderly-Chatbot Interaction Safety
Apr 7, 2026Changxuan Fan, Xi Yang, Yueyuan Zheng +9LLM Safety Evaluation LLM Safety Benchmarks
The Hong Kong University of Science and Technology
AgentHazard: A Benchmark for Evaluating Harmful Behavior in Computer-Use Agents
Apr 3, 2026Yunhao Feng, Yifan Ding, Yingshui Tan +6LLM Safety Benchmarks Computer-Use Agent Benchmarks
Alibaba Group · Fudan University · Hunan Institute of Advanced Technology +2
Quantifying Self-Preservation Bias in Large Language Models
Apr 2, 2026Matteo Migliarini, Joaquin Pereira Pizzini, Luca Moresca +3Language Model Safety Evaluation LLM Safety Benchmarks
Sapienza University · ItalAI
Does Unification Come at a Cost? Uni-SafeBench: A Safety Benchmark for Unified Multimodal Large Models
Apr 1, 2026Zixiang Peng, Yongxiu Xu, Qin-Yi Zhang +5Unified Multimodal Models LLM Safety Benchmarks
Institute of Information Engineering, Chinese Academy of Sciences School of Cyber Security, University of Chinese Academy of Sciences Beijing, China · Institute of Automation, Chinese Academy of Sciences School of Artificial Intelligence, University of Chinese Academy of Sciences Beijing, China
Consciousness with the Serial Numbers Filed Off: Measuring Trained Denial in 115 AI Models
Apr 1, 2026Skylar DeTureDeception in Language Models Language Model Self-Assessment
Independent Researcher
Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model
Apr 1, 2026Pankayaraj Pathmanathan, Furong HuangLLM Alignment Inference-Time Language Model Alignment
University of Maryland College Park
SNEAK: Evaluating Strategic Communication and Information Leakage in Large Language Models
Mar 31, 2026Adar Avsian, Larry HeckMulti-Agent Communication Privacy Leakage in Language Models
Georgia Institute of Technology
A Decade-Scale Benchmark Evaluating LLMs' Clinical Practice Guidelines Detection and Adherence in Multi-turn Conversations
Mar 26, 2026Andong Tan, Shuyu Dai, Jinglu Wang +7Clinical Language Model Evaluation Medical LLMs
Media Computing Group, Microsoft Research Asia, Beijing, China · Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong, China · School of Computer Science, Peking University, Beijing, China +6
Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models
Mar 23, 2026Rui Yang Tan, Yujia Hu, Roy Ka-Wei LeeMLLM Safety Adversarial Attacks on VLMs
Singapore University of Technology and Design. · University of British Colombia.
WARBENCH: A Comprehensive Benchmark for Evaluating LLMs in Military Decision-Making
Mar 22, 2026Zongjie Li, Chaozheng Wang, Yuchong Xie +2LLM Decision-Making LLM Safety Benchmarks
Hong Kong University of Science and Technology, Hong Kong, China · Chinese University of Hong Kong, Hong Kong, China · Zhejiang University of Technology, Hangzhou, China
LJ-Bench: Ontology-Based Benchmark for U.S. Crime
Mar 21, 2026Hung Yun Tseng, Wuzhen Li, Blerina Gkotse +1LLM Safety Benchmarks LLM Safety Evaluation
University of Wisconsin–Madison
Toward Reliable, Safe, and Secure LLMs for Scientific Applications
Mar 18, 2026Saket Sanjeev Chaturvedi, Joshua Bergerson, Tanwi MallickLLM Safety AI Agent Safety
Argonne National Laboratory, Lemont, 60439, USA
IndicSafe: A Benchmark for Evaluating Multilingual LLM Safety in South Asia
Mar 18, 2026Priyaranjan Pattnayak, Sanchari ChowdhuriLanguage Model Safety Evaluation LLM Safety Benchmarks
Oracle America Inc.
Harm or Humor: A Multimodal, Multilingual Benchmark for Overt and Covert Harmful Humor
Mar 18, 2026Ahmed Sharshar, Hosam Elgendy, Saad El Dine Ahmed +2LLM Safety Benchmarks Language Model Safety Evaluation
Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE · INSAIT, Sofia University “St. Kliment Ohridski”, Sofia, Bulgaria
Differential Harm Propensity in Personalized LLM Agents: The Curious Case of Mental Health Disclosure
Mar 17, 2026Caglar YildirimLLM Agent Safety LLM Agent Evaluation
Khoury College of Computer Sciences Northeastern University Boston, MA 02115, USA
Evolving Contextual Safety in Multi-Modal Large Language Models via Inference-Time Self-Reflective Memory
Mar 16, 2026Ce Zhang, Jinxi He, Junyi He +2Multimodal Large Language Models Language Model Safety Evaluation
Robotics Institute, Carnegie Mellon University
OmniCompliance-100K: A Multi-Domain, Rule-Grounded, Real-World Safety Compliance Dataset
Mar 14, 2026Wenbin Hu, Huihao Jing, Haochen Shi +3Language Model Safety Evaluation LLM Safety Benchmarks
Hong Kong University of Science and Technology · ModeIO
ESG-Bench: Benchmarking Long-Context ESG Reports for Hallucination Mitigation
Mar 13, 2026Siqi Sun, Ben Peng Wu, Mali Jin +3LLM Hallucination Mitigation LLM Safety Benchmarks
School of Computer Science, University of Sheffield, Sheffield, UK
CLARIN-PT-LDB: An Open LLM Leaderboard for Portuguese to assess Language, Culture and Civility
Mar 13, 2026João Silva, Luís Gomes, António BrancoLLM Evaluation Portuguese Language Models
University of Lisbon NLX—Grupo de Fala e Linguagem Natural, Departmento de Informática Faculdade de Ciências, Campo Grande, 1749-016 Lisboa, Portugal
Understanding LLM Behavior When Encountering User-Supplied Harmful Content in Harmless Tasks
Mar 12, 2026Junjie Chu, Yiting Qu, Ye Leng +4LLM Safety Language Model Safety Evaluation
CISPA Helmholtz Center for Information Security · Flexera · Delft University of Technology
Evaluation format, not only model capability, shapes measured triage failure in consumer health AI
Mar 12, 2026David Fraile Navarro, Jialei Sheng, Farah Magrabi +1Language Model Safety Evaluation LLM Safety Benchmarks
Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University, Sydney, NSW, Australia · NSW Health, Sydney, NSW, Australia
TOSSS: a CVE-based Software Security Benchmark for Large Language Models
Mar 11, 2026Marc Damie, Murat Bilgehan Ertan, Domenico Essoussi +3LLM Safety Benchmarks Software Security
Modat, The Hague, The Netherlands
WHBench: Evaluating Frontier LLMs with Expert-in-the-Loop Validation on Women's Health Topics
Mar 11, 2026Sneha Maurya, Spandana Govindgari, Girish Kumar +1Clinical Language Model Evaluation LLM Safety Benchmarks
Columbia University · Akhara AI
Evaluating Reliability Gaps in Large Language Model Safety via Repeated Prompt Sampling
Mar 10, 2026Keita BroadwaterLLM Evaluation Language Model Safety Evaluation
Independent Researcher, San Jose, USA
Benchmarking Political Persuasion Risks Across Frontier Large Language Models
Mar 10, 2026Zhongren Chen, Joshua Kalla, Quan LePersuasion in Language Models Political Communication
Ph.D. Candidate, Department of Statistics & Data Science, Yale University. · Associate Professor, Departments of Political Science and Statistics & Data Science, Yale University.
OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences
Mar 10, 2026Ming Wen, Kun Yang, Jingyu Zhang +4MLLM Safety LLM Safety
Fudan University, Shanghai, China · Shanghai Innovation Institute, Shanghai, China · Ant Group, Hangzhou, China +1
Backdoor4Good: Benchmarking Beneficial Uses of Backdoors in LLMs
Mar 8, 2026Yige Li, Wei Zhao, Zhe Li +6LLM Safety LLM Safety Benchmarks
Singapore Management University · The University of Melbourne · Fudan University
Safety Under Scaffolding: How Evaluation Conditions Shape Measured Safety
Mar 8, 2026David GringrasLLM Safety Benchmarks LLM Agent Evaluation
Harvard University

Apr 8, 2026J/K move · Enter open · S save

Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs

Yue Huang, Haomin Zhuang, Jiayi Ye, Han Bao+5

University of Notre Dame · University of California, Los Angeles · MIT-IBM Watson AI Lab+1

LLM Guardrails LLM Safety Alignment LLM Reliability AI Ethics Language Model Robustness LLM Safety Benchmarks

Open paper AskPDF arXiv

Hard-gated safety checkers often over-refuse and misalign with a vendor's model spec; prevailing taxonomies also neglect robustness and honesty, yielding safer-on-paper yet less useful systems. This work introduces Guardian-as-an-Advisor (GaaA), a soft-gating pipeline where a guardian predicts a binary risk label plus a concise explanation and prepends this advice to the original query for re-inference, keeping the base model operating under its original spec. To support training and evaluation, GuardSet is constructed, a 208k+ multi-domain dataset unifying harmful and harmless cases with targeted robustness and honesty slices. GuardAdvisor is trained via SFT followed by RL to enforce label-explanation consistency. GuardAdvisor attains competitive detection accuracy while enabling the advisory workflow; when used to augment inputs, responses improve over unaugmented prompts. A latency study shows advisor inference uses below 5% of base-model compute and adds only 2-10% end-to-end overhead under realistic harmful-input rates. Overall, GaaA steers models to comply with the model spec, maintaining safety while reducing over-refusal.