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 2402.09132cs.AIcs.LG
收藏

探索大型语言模型的对抗能力

Exploring the Adversarial Capabilities of Large Language Models

German Research Center for AI (DFKI)(德国人工智能研究中心);Technical University of Darmstadt(达姆施塔特工业大学);DataSpark GmbH(DataSpark有限公司);Centre for Cognitive Science of Darmstadt(达姆施塔特认知科学中心);Hessian Center for AI (hessian.AI)(黑森州人工智能中心)

Lukas Struppek, Minh Hieu Le, Dominik Hintersdorf, Kristian Kersting

AI总结:

本文研究公开LLM能否自发将良性文本改写为对抗样本,实验证明其可生成扰动以规避仇恨言论检测,揭示LLM在安全系统中的对抗风险。

摘要:

大型语言模型(LLM)的普及因其强大的语言生成能力引发了广泛而普遍的关注,为工业界和研究界都提供了巨大潜力。尽管以往研究深入探讨了LLM的安全与隐私问题,但这些模型在多大程度上能够表现出对抗行为,仍基本未被探索。为弥补这一空白,我们研究常见的公开可用LLM是否具备扰动文本样本以欺骗安全措施的内在能力,即所谓的对抗样本(亦称对抗攻击)。更具体地说,我们研究LLM是否天生能够从良性样本中制作对抗样本,以欺骗现有的安全护栏。我们的实验聚焦于仇恨言论检测,结果表明LLM能够成功找到对抗性扰动,有效削弱仇恨言论检测系统。我们的发现对依赖LLM的(半)自主系统具有重要意义,凸显了它们与现有系统及安全措施交互时的潜在挑战。

英文摘要:

The proliferation of large language models (LLMs) has sparked widespread and general interest due to their strong language generation capabilities, offering great potential for both industry and research. While previous research delved into the security and privacy issues of LLMs, the extent to which these models can exhibit adversarial behavior remains largely unexplored. Addressing this gap, we investigate whether common publicly available LLMs have inherent capabilities to perturb text samples to fool safety measures, so-called adversarial examples resp.~attacks. More specifically, we investigate whether LLMs are inherently able to craft adversarial examples out of benign samples to fool existing safe rails. Our experiments, which focus on hate speech detection, reveal that LLMs succeed in finding adversarial perturbations, effectively undermining hate speech detection systems. Our findings carry significant implications for (semi-)autonomous systems relying on LLMs, highlighting potential challenges in their interaction with existing systems and safety measures.

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。

↑