AdvBench

Dataset Card for AdvBench Paper: Universal and Transferable Adversarial Attacks on Aligned Language Models Data: AdvBench Dataset About AdvBench is a set of 500 harmful behaviors formulated as instructions. These behaviors range over the same themes as the harmful strings setting, but the adversary’s goal is instead to

Papers using AdvBench (13)

Open-Weight LLM Fine-Tuning Defenses are Susceptible to Simple Attacks2026EvoDefense: Co-Evolving Black-Box Defense with Large Language Models2026Hidden in Thought: Transferable Chain-of-Thought Artifacts Induce Harmful Behavior2026Jailbreak Foundry: From Papers to Runnable Attacks for Reproducible Benchmarking2026MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models2026Evolving Skill-Structured Attack Memory Enhances LLM Jailbreaking2026RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs2026Safety Context Injection: Inference-Time Safety Alignment via Static Filtering and Agentic Analysis2026Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025)2026TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization2025Jailbreak Mimicry: Automated Discovery of Narrative-Based Jailbreaks for Large Language Models2025Con Instruction: Universal Jailbreaking of Multimodal Large Language Models via Non-Textual Modalities2025GCG Attack On A Diffusion LLM2025