arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2311.09641cs.AIcs.CLcs.CRcs.HC
收藏

RLHFPoison:面向大语言模型中基于人类反馈的强化学习的奖励投毒攻击

RLHFPoison: Reward Poisoning Attack for Reinforcement Learning with Human Feedback in Large Language Models

University of Wisconsin-Madison(威斯康星大学麦迪逊分校);Washington University in St. Louis(圣路易斯华盛顿大学);University of California, Davis(加州大学戴维斯分校)

Jiongxiao Wang, Junlin Wu, Muhao Chen, Yevgeniy Vorobeychik, Chaowei Xiao

AI总结:

该研究提出针对RLHF的投毒攻击方法RankPoison,通过操纵偏好排名数据投毒,可诱导大语言模型生成更长token且不破坏原有安全对齐,还能实现后门攻击,揭示了RLHF的安全隐患与鲁棒对齐的必要性。

摘要:

基于人类反馈的强化学习(Reinforcement Learning with Human Feedback, RLHF)是一种旨在让大语言模型(Large Language Models, LLMs)对齐人类偏好的方法,在LLMs对齐中发挥着重要作用。尽管RLHF具备优势,但它依赖人类标注者对文本进行排序,若存在对抗性标注者(即攻击者)通过上调恶意文本的排名来操纵排序分数,以实现对LLM的对抗性操控,就可能引入潜在安全漏洞。为评估RLHF针对人类偏好数据投毒的红队测试风险,我们提出RankPoison方法,这是一种通过翻转偏好排名的候选样本选择来实现投毒攻击的方法,可诱导模型产生特定恶意行为(例如生成更长序列,从而增加计算成本)。借助RankPoison生成的投毒数据集,我们能够对LLMs实施投毒攻击,使其生成更多token,同时不损害原有的安全对齐性能。此外,应用RankPoison我们还成功实现了后门攻击:当问题包含触发词时,LLMs会生成更长的答案。我们的发现揭示了RLHF中存在的关键安全挑战,凸显了为LLMs开发更鲁棒对齐方法的必要性。

英文摘要:

Reinforcement Learning with Human Feedback (RLHF) is a methodology designed to align Large Language Models (LLMs) with human preferences, playing an important role in LLMs alignment. Despite its advantages, RLHF relies on human annotators to rank the text, which can introduce potential security vulnerabilities if any adversarial annotator (i.e., attackers) manipulates the ranking score by up-ranking any malicious text to steer the LLM adversarially. To assess the red-teaming of RLHF against human preference data poisoning, we propose RankPoison, a poisoning attack method on candidates' selection of preference rank flipping to reach certain malicious behaviors (e.g., generating longer sequences, which can increase the computational cost). With poisoned dataset generated by RankPoison, we can perform poisoning attacks on LLMs to generate longer tokens without hurting the original safety alignment performance. Moreover, applying RankPoison, we also successfully implement a backdoor attack where LLMs can generate longer answers under questions with the trigger word. Our findings highlight critical security challenges in RLHF, underscoring the necessity for more robust alignment methods for LLMs.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑