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 2411.01344cs.HCcs.AIcs.CR
收藏

被AI观点掩盖的隐私泄露:语言模型智能体中人类隐私监督的研究

Privacy Leakage Overshadowed by Views of AI: A Study on Human Oversight of Privacy in Language Model Agent

Northeastern University(东北大学);University of Washington(华盛顿大学)

Zhiping Zhang, Bingcan Guo, Tianshi Li

AI总结:

该研究通过300人任务调查发现,人类监督语言模型智能体时易忽视隐私泄露,使有害披露占比从15.7%升至55%,还识别出六种隐私行为模式,为隐私保护型智能体系统设计提供参考。

摘要:

代表用户执行个人任务(如回复邮件)的语言模型(LM)智能体可提升生产力,但也易出现意外的隐私泄露风险。我们开展了首项关于人类监督LM智能体隐私影响能力的研究。通过一项基于任务的调查(样本量N=300),我们探究了在异步人际沟通任务中,相较于自己撰写的回复,人们对LM智能体生成的回复的反应与评估方式。我们发现,人们可能更青睐隐私泄露程度更高的智能体回复,而非自己起草的回复,或是认为两者都不错,这导致有害信息披露占比从15.7%升至55.0%。我们进一步识别出六种隐私行为模式,反映了人们在监督LM智能体行为时背后不同的顾虑程度、信任水平与隐私偏好。我们的研究结果为设计可实现隐私保护交互、并就隐私偏好达成双向对齐的智能体系统提供了思路,以帮助用户校准信任。

英文摘要:

Language model (LM) agents that act on users' behalf for personal tasks (e.g., replying emails) can boost productivity, but are also susceptible to unintended privacy leakage risks. We present the first study on people's capacity to oversee the privacy implications of the LM agents. By conducting a task-based survey ($N=300$), we investigate how people react to and assess the response generated by LM agents for asynchronous interpersonal communication tasks, compared with a response they wrote. We found that people may favor the agent response with more privacy leakage over the response they drafted or consider both good, leading to an increased harmful disclosure from 15.7% to 55.0%. We further identified six privacy behavior patterns reflecting varying concerns, trust levels, and privacy preferences underlying people's oversight of LM agents' actions. Our findings shed light on designing agentic systems that enable privacy-preserving interactions and achieve bidirectional alignment on privacy preferences to help users calibrate trust.

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。

↑