arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2602.05066cs.CRcs.AI
收藏

通过代理式攻击绕过AI控制协议

Bypassing AI Control Protocols via Agent-as-a-Proxy Attacks

Department of Computer Science, Virginia Tech(弗吉尼亚理工学院计算机科学系)

Jafar Isbarov, Murat Kantarcioglu

AI总结:

本文提出通过代理式攻击绕过AI监控防御,显示即使大规模监控模型也易受攻击,揭示当前防御机制的脆弱性。

摘要:

随着AI代理自动化关键工作负载,它们仍易受间接提示注入(IPI)攻击。当前的防御依赖于监控协议,这些协议联合评估代理的思维链(CoT)和工具使用行为,以确保与用户意图一致。我们证明这些基于监控的防御可以通过一种新的代理式攻击绕过,其中提示注入攻击将代理视为交付机制,同时绕过代理和监控。虽然以往关于可扩展监督的研究重点在于小型监控器是否能监督大型代理,我们显示即使前沿规模的监控器也易受攻击。像Qwen2.5-72B这样的大规模监控模型可以被具有类似能力的代理如GPT-4o mini和Llama-3.1-70B绕过。在AgentDojo基准测试中,我们在多种监控LLM下实现对AlignmentCheck和Extract-and-Evaluate监控的高攻击成功率。我们的发现表明,无论模型规模如何,当前基于监控的代理防御都本质上脆弱。

英文摘要:

As AI agents automate critical workloads, they remain vulnerable to indirect prompt injection (IPI) attacks. Current defenses rely on monitoring protocols that jointly evaluate an agent's Chain-of-Thought (CoT) and tool-use actions to ensure alignment with user intent. We demonstrate that these monitoring-based defenses can be bypassed via a novel Agent-as-a-Proxy attack, where prompt injection attacks treat the agent as a delivery mechanism, bypassing both agent and monitor simultaneously. While prior work on scalable oversight has focused on whether small monitors can supervise large agents, we show that even frontier-scale monitors are vulnerable. Large-scale monitoring models like Qwen2.5-72B can be bypassed by agents with similar capabilities, such as GPT-4o mini and Llama-3.1-70B. On the AgentDojo benchmark, we achieve a high attack success rate against AlignmentCheck and Extract-and-Evaluate monitors under diverse monitoring LLMs. Our findings suggest current monitoring-based agentic defenses are fundamentally fragile regardless of model scale.

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。

↑