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 2610.18508cs.CRcs.LG
收藏

面向计算机使用的安全自主技术

Towards Safe Autonomy for Computer Use

Tübingen AI Center, University of Tübingen(蒂宾根大学蒂宾根人工智能中心)

Dominik Seip, Bartol Markovinović, Matthias Hein

AI总结:

本文针对计算机使用智能体(CUA)的安全问题,提出双LLM框架SA-CUA,通过构建信任边界实现自主处理不可信信息任务,其效用优于CaMeL-CUA且攻击成功率为零。

摘要:

基于大语言模型(LLM)的智能体当前的安全性依赖于对齐训练和安全分类器,但两者均无法阻止复杂的间接提示注入(IPI)绕过防御并造成重大危害。近期研究因此构建了具有信息流保障的智能体,这类智能体依赖结构化工具接口与环境交互。计算机使用智能体(CUA)会在计算机屏幕上执行视觉操作,而屏幕无法区分可信的第一方接口与可能由攻击者生成的不可信内容;操作本身也存在歧义,一次点击可能落在应用工具栏或攻击者控制的按钮上。将用户视为信任来源并确认每一项敏感决策会导致确认疲劳,进而可能遗漏有害操作。本文提出SA-CUA,这是一种双LLM框架,它通过从无障碍树推导保守确定性规则,从特权LLM的截图中揭示可信区域,从而为CUA生成可靠的信任边界,将可信与不可信内容分离。人类交互仅需在任务开始时用于正确指定任务;任务执行期间,一个稳健的背书通道允许通过可信通道验证信息,使值从不可信域传递到可信域,从而让智能体自主处理涉及不可信信息的任务。SA-CUA的效用远高于CaMeL-CUA,同时将攻击成功率保持为零。

英文摘要:

The safety of LLM-based agents currently rests on alignment training and safety classifiers, neither of which can rule out that a sophisticated indirect prompt injection (IPI) goes through and causes significant harm. Recent work therefore builds agents with information-flow guarantees that rely on structured tool interfaces to interact with the environment. Computer Use Agents (CUAs) act visually on the computer screen, which does not provide a boundary between trusted first-party interfaces and untrusted, potentially attacker-authored content. Actions are equally ambiguous: a click may land on an application toolbar or a button controlled by an attacker. Treating the user as the source of trust and confirming each sensitive decision produces confirmation fatigue, with the risk of missing harmful actions. We introduce SA-CUA, a dual-LLM framework that generates a reliable trust boundary for CUAs by deriving conservative deterministic rules from the accessibility tree for unmasking trusted regions from the privileged LLM's screenshot, keeping trusted and untrusted content separated. Human interaction is only required for proper task specification in the beginning. During the task a robust endorsement channel allows values to pass from the untrusted to the trusted domain by verifying information via trusted channels. This allows the agent to autonomously handle tasks involving untrusted information. SA-CUA maintains much higher utility than CaMeL-CUA while keeping the attack success rate at zero.

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。

↑