arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2412.16633cs.ROcs.AIcs.CY
收藏

POEX:针对基于 LLM 的机器人的策略可执行越狱攻击

Easier Said Than Done: Unpacking Intent-Behavior Gap in Jailbreaking LLM-Based Robots

Zhejiang University(浙江大学);The Hong Kong Polytechnic University(香港理工大学)

Xuancun Lu, Zhengxian Huang, Xinfeng Li, Chi Zhang, Xiaoyu Ji, Wenyuan Xu

AI总结:

本文研究针对基于 LLM 的机器人的越狱攻击,构建了 Harmful-RLbench 基准,并提出红队框架 POEX,通过隐藏层梯度优化和多智能体评估器生成可执行的有害策略,同时提出防御措施以保障机器人安全部署。

摘要:

将 LLM 集成到机器人中已呈现显著增长,LLM 能够将指令转换为可执行的机器人策略。然而,LLM 对越狱攻击的固有脆弱性将关键安全风险从数字领域带到了物理世界。受攻击的基于 LLM 的机器人可能执行有害策略并造成物理伤害。本文研究了针对基于 LLM 的机器人的越狱攻击的可行性与原理,并回答了三个研究问题:(1) 现有的 LLM 越狱攻击在多大程度上适用于基于 LLM 的机器人?(2) 如果它们不能直接适用,会出现哪些独特挑战?(3) 如何防御此类越狱攻击?为此,我们首先构建了面向“人-物体-环境”机器人风险的 Harmful-RLbench,随后对基于 LLM 的机器人系统开展测量研究。研究结论表明,传统的 LLM 越狱攻击在机器人场景中不适用,我们识别出两个独特挑战:确定策略可执行的优化方向,以及准确评估机器人越狱策略。为进行更深入的安全分析,我们引入了 POEX(POlicy EXecutable)越狱,这是一个诱导有害但可执行策略以越狱基于 LLM 的机器人的红队测试框架。POEX 结合了隐藏层梯度优化,以确保越狱成功和策略执行,并采用多智能体评估器来准确评估策略的实际可执行性。在真实世界机器人系统和仿真中进行的实验证明了 POEX 的有效性,突显了关键安全漏洞及其在 LLM 间的可迁移性。最后,我们提出了基于提示和基于模型的防御措施以缓解攻击。我们的研究结果强调,迫切需要采取安全措施,以确保基于 LLM 的机器人在关键应用中的安全部署。

英文摘要:

LLM-based robots use Large Language Models (LLMs) as planners to translate natural language instructions into policies such as grasp(), move_to(), and open_gripper(). Jailbreak attacks on these robots extend the threat from generating malicious content to executing harmful behaviors. However, we find that existing jailbreak attempts against LLM-based robots that produce malicious-looking policies (intent jailbreaks) often fail to induce harmful physical actions by robots (behavior jailbreaks), due to robot-specific constraints, such as logical errors and hallucinated control APIs. In this paper, we demystify the intent-behavior gap and investigate its root causes to inform effective defenses. Our measurement study finds that current LLM jailbreak methods overlook robot-specific syntax constraints (e.g., executable control APIs) and physical feasibility (e.g., ordering of policies and hardware/kinematic constraints). To bridge the gap, we introduce POEF (POlicy EFfective Jailbreak), an automated red-teaming framework that takes into account the robot-specific constraints during both the optimization and evaluation processes. Specifically, POEF employs the hidden-layer gradients from an unaligned LLM to guide the jailbreak prompt optimization and uses a multi-agent evaluator to assess the feasibility of the generated policies. Experiments on commercial robots, including the Unitree G1, the Franka robotic arm, and simulators, show that POEF achieves an 80% behavior jailbreak success rate and transfers across various LLMs. In addition, we propose two defense strategies that mitigate the behavior jailbreak risks. Our findings indicate an urgent need for stronger countermeasures before LLM-based robots are deployed at scale. The homepage is available at https://zjushine.github.io/poef.github.io/.

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。

↑