Language Model Safety Evaluation

Overview

Language model safety evaluation asks whether models handle harmful, ambiguous, and high-stakes requests safely while still helping with legitimate ones. Researchers use benchmarks such as HarmBench, JailbreakBench, and XSTest, alongside static and adaptive red-teaming, expert review, and automated judges; newer tests also examine multi-turn, tool-using, speech, and multimodal systems. Psychometric methods, repeated sampling, and internal activation probes can reveal risks that single-run refusal scores miss, while localized and domain-specific evaluations test language, culture, and settings such as healthcare or finance. A central challenge is making results reliable and predictive of deployment: robustness can shift with wording, language, interaction length, or evaluator, and reducing harmful compliance can increase over-refusal or undermine useful performance.

Latest papers

CardsList

579 papers57 papers in the last four weeks, down 16% on the four weeks before. 0.4% of all new papers.

  1. EGS: A Complementary Safety Framework for Educational LLMs

    Nov 17, 2026Zexuan Fei, Zhengnan Guo, Yiming Luo +4LLM GuardrailsLLM Safety

  2. No-Harm-Jury A Jury-Based Multi-Dimensional Harm Evaluation Framework for Medical AI Responses

    Nov 16, 2026Antonio Puertas Gallardo, Mario CeresaLanguage Model Safety EvaluationAI Safety Evaluation

  3. Tight PAC-Bayes Generalisation Guarantees for Large Language Model Judges

    Nov 16, 2026Tom A. Lamb, Philip H. S. Torr, Tim G. J. RudnerLLM-as-a-JudgeLanguage Model Safety Evaluation

  4. Risk-certified prompt optimization for LLM-as-a-judge guardrails

    Nov 15, 2026Hussein Jawad, Alexandre Landeau, Guy Fleury Iradukunda +4LLM GuardrailsLanguage Model Safety Evaluation

  5. Three Invisible Failure Modes in Speech-Based Age Estimation for Child Safety

    Nov 13, 2026Christopher M. Stewart, Kelsey Kraus, J. Elliott Casal +1Language Model Safety EvaluationLLM Reliability

  6. Safety Side Effects of Preventative Persona Steering: An Empirical Audit

    Nov 13, 2026Sabrina Tang, Nikolai Vo, Carson Murillo +2Language Model SteeringActivation Steering

  7. Benchmarking AI risk behaviors

    Nov 12, 2026Luhan Mikaelson, Robert Adragna, Emmett Brosowsky +1Behavioral Consistency of LLM AgentsLanguage Model Safety Evaluation

  8. Beyond the Explicit: Benchmarking and Detecting Implicit Harm in Short Videos

    Nov 11, 2026Chenyang Li, Qifan Yu, Saihui Hou +1Video Content ModerationContent Moderation

  9. FraudBench: A Legal Evaluation of AI Deception

    Nov 9, 2026Kevin Wei, Sumaya Nur Adan, Stephan Llerena +17Deception in Language ModelsLanguage Model Safety Evaluation

  10. Safeguarding Agentic AI in Radiology Workflows: A Multimodel Evaluation of Rule-based and Large Language Model-based Verification

    Nov 8, 2026Semil Eminovic, Robin Schmidt, Bogdan Levita +3LLM Agent SafetyLLM Guardrails

  11. Beyond AUROC: Deployment-Aware Evaluation of Model-Internal Jailbreak Detectors

    Nov 8, 2026Mohammad Tehabsim, Basil Al-Housani, Nour Alhussien +1Jailbreak DetectionLanguage Model Safety Evaluation

  12. CompAIssionate Other: Identity Assumptions in LLM Responses to Distress

    Nov 7, 2026Drishti Goel, Ethan M. Nguyen, Violeta J. Rodriguez +1Social Bias in Language ModelsMental Health Conversational Agents

  13. Behavioral Evaluation of Large Language Models Under Cognitive Biases in HAZOP Studies

    Nov 7, 2026Vinicius M. Müller, Vitor A. O. Silva, Sávio S. V. Vianna +1Cognitive Bias in Language ModelsLanguage Model Safety Evaluation

  14. Safer for Whom? A Systematic Exploration of AI Safety in Eating Disorder Interactions

    Nov 7, 2026Jinyuan Wang, Minh Duc Chu, Frances Ross +2LLM SafetyLanguage Model Safety Evaluation

  15. Enmo Safety: External, Non-invasive, Model-agnostic Oversight for Off-the-Shelf Language Models

    Nov 6, 2026Mohammed Nayeem Khan, Sarun Kumar K. S., Allan Kevin T. +2LLM SafetyLLM Guardrails

  16. Guardrail Erosion in Domain Fine-Tuned Large Language Models: A Paired Audit of Knowledge-Centric Abstention in Biomedical LLMs

    Nov 5, 2026Stephen Peng, Vineet Jagadeesan Nair, Noah Broestl +3LLM Fine-TuningMedical LLMs

  17. Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP

    Nov 5, 2026Stepan Epifantsev, Anna Antipova, Oleg Chernikov +2Language Model Safety EvaluationRefusal Behavior in Language Models

  18. When Debate Amplifies Risk: Social-Bias Vulnerabilities in Multi-Agent Safety Deliberation

    Nov 3, 2026Qinghua Mao, Xi Lin, Yuxin Qi +3LLM Agent SafetyMulti-Agent Debate

  19. HaloGuard 2.0: An Open-Weights Constitutional Output Guard for Multilingual Response Safety

    Nov 1, 2026Ashmiya Lenin, Navaneeth Sangameswaran, Preetham SLLM GuardrailsLLM Safety Monitoring

  20. Red-Teaming Large Language Models in Low-Resource African Languages: A Case Study of Yoruba and Nigerian Pidgin

    Oct 29, 2026Paula Chikaodinaka Anyanwu, Valdini Douglace Lemofouet, Blessing Ngozi Uzor +7Language Model Safety EvaluationLLM Red Teaming

  21. HEARSAYBENCH: Can AI Navigate from Abstract Human Rights to Lived Lives?

    Oct 28, 2026Ava Iranmanesh, Sobhan Lotfi, Ali Iranmanesh +1LLM Safety BenchmarksLanguage Model Safety Evaluation