HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Center for AI Safety（人工智能安全中心）；Carnegie Mellon University（卡内基梅隆大学）；UC Berkeley（加州大学伯克利分校）；Microsoft（微软）
Mantas Mazeika, Long Phan, Xuwang Yin, Andy Zou, Zifan Wang, Norman Mu, Elham Sakhaee, Nathaniel Li, Steven Basart, Bo Li, David Forsyth, Dan Hendrycks
Automated red teaming holds substantial promise for uncovering and mitigating the risks associated with the malicious use of large language models (LLMs), yet the field lacks a standardized evaluation framework to rigorously assess new methods. To address this issue, we introduce HarmBench, a standardized evaluation framework for automated red teaming. We identify several desirable properties previously unaccounted for in red teaming evaluations and systematically design HarmBench to meet these criteria. Using HarmBench, we conduct a large-scale comparison of 18 red teaming methods and 33 target LLMs and defenses, yielding novel insights. We also introduce a highly efficient adversarial training method that greatly enhances LLM robustness across a wide range of attacks, demonstrating how HarmBench enables codevelopment of attacks and defenses. We open source HarmBench at https://github.com/centerforaisafety/HarmBench.
University of Chinese Academy of Sciences（中国科学院大学）；Inner Mongolia University of Technology（内蒙古科技大学）；Tsinghua University（清华大学）；Shandong University（山东大学）
Graph-Regularized Sparse Autoencoders for LLM Safety Steering
ELLIS Institute Tübingen（图宾根ELLIS研究所）；Max Planck Institute for Intelligent Systems（智能系统马克斯·普朗克研究所）；Intesa Sanpaolo（Intesa Sanpaolo公司）；University of Southern California（南加州大学）
REPRO: Reflective Evolutionary Prompt Optimizer for Automated LLM Red-Teaming
King Fahd University of Petroleum & Minerals（法赫德国王石油与矿产大学）；Imam Abdulrahman Bin Faisal University（伊玛目阿卜杜勒拉赫曼·本·费萨尔大学）；SDAIA-KFUPM Joint Research Center for Artificial Intelligence（SDAIA-KFUPM人工智能联合研究中心）
Lapis Labs（拉皮斯实验室）；University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Center for AI Safety（AI安全中心）；University of California, San Diego（加利福尼亚大学圣迭戈分校）；Gray Swan AI（格雷天鹅人工智能公司）；Carnegie Mellon University（卡内基梅隆大学）；Harvard University（哈佛大学）；University of California, Berkeley（加利福尼亚大学伯克利分校）；University of Chicago（芝加哥大学）
