arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.15001cs.LG
收藏

黑盒大语言模型的边界点劫持

Boundary Point Jailbreaking of Black-Box LLMs

UK AI Security Institute(英国人工智能安全研究所);OATML, University of Oxford(OATML、牛津大学)

Xander Davies, Giorgi Giglemiani, Edmund Lau, Eric Winsor, Geoffrey Irving, Yarin Gal

AI总结:

BPJ是一种全新的黑盒自动化劫持攻击方法,通过边界点选择策略有效突破宪法分类器和GPT-5输入分类器的防护。

摘要:

前沿大语言模型通过对抗性提示(称为'劫持')来防止提取有害信息。最近,防御者开发了基于分类器的系统,已通过数千小时的人类红队测试。我们引入了边界点劫持(BPJ),一种新的自动化劫持攻击方法,能够逃避最强的行业部署防护。与以往依赖白盒/灰盒假设(如分类器分数或梯度)或现有劫持库的攻击不同,BPJ是完全黑盒的,每个查询仅使用一个信息位:分类器是否标记交互。为此,BPJ解决了在对抗鲁棒现实防御中优化攻击的核心难题:评估所提出攻击修改是否改进。而不是直接学习针对目标有害字符串的攻击,BPJ将字符串转换为一系列中间攻击目标课程,然后主动选择最佳检测攻击强度微小变化的'边界点'。我们相信BPJ是首个成功开发针对宪法分类器的通用劫持攻击算法,也是首个无需依赖人类攻击种子即可成功攻击GPT-5输入分类器的自动化攻击算法。BPJ在单次交互中难以防御,但在优化过程中会触发大量标志,表明有效的防御需要补充单次交互方法与批量级监控。

英文摘要:

Frontier LLMs are safeguarded against attempts to extract harmful information via adversarial prompts known as "jailbreaks". Recently, defenders have developed classifier-based systems that have survived thousands of hours of human red teaming. We introduce Boundary Point Jailbreaking (BPJ), a new class of automated jailbreak attacks that evade the strongest industry-deployed safeguards. Unlike previous attacks that rely on white/grey-box assumptions (such as classifier scores or gradients) or libraries of existing jailbreaks, BPJ is fully black-box and uses only a single bit of information per query: whether or not the classifier flags the interaction. To achieve this, BPJ addresses the core difficulty in optimising attacks against robust real-world defences: evaluating whether a proposed modification to an attack is an improvement. Instead of directly trying to learn an attack for a target harmful string, BPJ converts the string into a curriculum of intermediate attack targets and then actively selects evaluation points that best detect small changes in attack strength ("boundary points"). We believe BPJ is the first fully automated attack algorithm that succeeds in developing universal jailbreaks against Constitutional Classifiers, as well as the first automated attack algorithm that succeeds against GPT-5's input classifier without relying on human attack seeds. BPJ is difficult to defend against in individual interactions but incurs many flags during optimisation, suggesting that effective defence requires supplementing single-interaction methods with batch-level monitoring.

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。

↑