迫使大语言模型做出并泄露(几乎)任何事
Coercing LLMs to do and reveal (almost) anything
AI总结:
本文系统梳理并分类了针对大语言模型的对抗性攻击,揭示其攻击面远不止越狱,并通过受控实验发现许多攻击源于编码预训练和词表中的故障token。
摘要:
近期研究表明,针对大语言模型(LLMs)的对抗性攻击可以“越狱”模型,使其生成有害言论。本文认为,针对LLMs的对抗性攻击范围远不止越狱。我们对可能的攻击面和攻击目标进行了广泛概述。基于一系列具体示例,我们讨论、分类并系统化了能够迫使模型产生各种非预期行为的攻击,例如误导、模型控制、拒绝服务或数据提取。我们在受控实验中分析了这些攻击,发现其中许多攻击源于预训练LLMs时赋予其编码能力的做法,以及常见LLM词表中持续存在奇怪的“故障”token,出于安全原因应将其移除。
英文摘要:
It has recently been shown that adversarial attacks on large language models (LLMs) can "jailbreak" the model into making harmful statements. In this work, we argue that the spectrum of adversarial attacks on LLMs is much larger than merely jailbreaking. We provide a broad overview of possible attack surfaces and attack goals. Based on a series of concrete examples, we discuss, categorize and systematize attacks that coerce varied unintended behaviors, such as misdirection, model control, denial-of-service, or data extraction. We analyze these attacks in controlled experiments, and find that many of them stem from the practice of pre-training LLMs with coding capabilities, as well as the continued existence of strange "glitch" tokens in common LLM vocabularies that should be removed for security reasons.
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。