Sahra Ghalebikesabi, Eugene Bagdasaryan, Ren Yi, Itay Yona, Ilia Shumailov, Aneesh Pappu, Chongyang Shi, Laura Weidinger, Robert Stanforth, Leonard Berrada, Pushmeet Kohli, Po-Sen Huang, Borja Balle
{"title":"在具有隐私意识的助手中操作情境完整性","authors":"Sahra Ghalebikesabi, Eugene Bagdasaryan, Ren Yi, Itay Yona, Ilia Shumailov, Aneesh Pappu, Chongyang Shi, Laura Weidinger, Robert Stanforth, Leonard Berrada, Pushmeet Kohli, Po-Sen Huang, Borja Balle","doi":"arxiv-2408.02373","DOIUrl":null,"url":null,"abstract":"Advanced AI assistants combine frontier LLMs and tool access to autonomously\nperform complex tasks on behalf of users. While the helpfulness of such\nassistants can increase dramatically with access to user information including\nemails and documents, this raises privacy concerns about assistants sharing\ninappropriate information with third parties without user supervision. To steer\ninformation-sharing assistants to behave in accordance with privacy\nexpectations, we propose to operationalize $\\textit{contextual integrity}$\n(CI), a framework that equates privacy with the appropriate flow of information\nin a given context. In particular, we design and evaluate a number of\nstrategies to steer assistants' information-sharing actions to be CI compliant.\nOur evaluation is based on a novel form filling benchmark composed of synthetic\ndata and human annotations, and it reveals that prompting frontier LLMs to\nperform CI-based reasoning yields strong results.","PeriodicalId":501479,"journal":{"name":"arXiv - CS - Artificial Intelligence","volume":"32 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-08-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Operationalizing Contextual Integrity in Privacy-Conscious Assistants\",\"authors\":\"Sahra Ghalebikesabi, Eugene Bagdasaryan, Ren Yi, Itay Yona, Ilia Shumailov, Aneesh Pappu, Chongyang Shi, Laura Weidinger, Robert Stanforth, Leonard Berrada, Pushmeet Kohli, Po-Sen Huang, Borja Balle\",\"doi\":\"arxiv-2408.02373\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Advanced AI assistants combine frontier LLMs and tool access to autonomously\\nperform complex tasks on behalf of users. While the helpfulness of such\\nassistants can increase dramatically with access to user information including\\nemails and documents, this raises privacy concerns about assistants sharing\\ninappropriate information with third parties without user supervision. To steer\\ninformation-sharing assistants to behave in accordance with privacy\\nexpectations, we propose to operationalize $\\\\textit{contextual integrity}$\\n(CI), a framework that equates privacy with the appropriate flow of information\\nin a given context. In particular, we design and evaluate a number of\\nstrategies to steer assistants' information-sharing actions to be CI compliant.\\nOur evaluation is based on a novel form filling benchmark composed of synthetic\\ndata and human annotations, and it reveals that prompting frontier LLMs to\\nperform CI-based reasoning yields strong results.\",\"PeriodicalId\":501479,\"journal\":{\"name\":\"arXiv - CS - Artificial Intelligence\",\"volume\":\"32 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-08-05\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - CS - Artificial Intelligence\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2408.02373\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - CS - Artificial Intelligence","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2408.02373","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
摘要
先进的人工智能助手结合了前沿 LLM 和工具访问,可代表用户自主执行复杂的任务。虽然这类助手在获取用户信息(包括电子邮件和文档)后能显著提高帮助性,但这也引发了隐私问题,即助手在没有用户监督的情况下与第三方共享不适当的信息。为了引导信息共享助手的行为符合隐私期望,我们提出了$\textit{contextual integrity}$(CI),这是一个将隐私等同于特定情境下适当信息流的框架。我们的评估基于一个由合成数据和人类注释组成的新颖的表单填写基准,它揭示了促使前沿 LLM 执行基于 CI 的推理会产生强大的结果。
Operationalizing Contextual Integrity in Privacy-Conscious Assistants
Advanced AI assistants combine frontier LLMs and tool access to autonomously
perform complex tasks on behalf of users. While the helpfulness of such
assistants can increase dramatically with access to user information including
emails and documents, this raises privacy concerns about assistants sharing
inappropriate information with third parties without user supervision. To steer
information-sharing assistants to behave in accordance with privacy
expectations, we propose to operationalize $\textit{contextual integrity}$
(CI), a framework that equates privacy with the appropriate flow of information
in a given context. In particular, we design and evaluate a number of
strategies to steer assistants' information-sharing actions to be CI compliant.
Our evaluation is based on a novel form filling benchmark composed of synthetic
data and human annotations, and it reveals that prompting frontier LLMs to
perform CI-based reasoning yields strong results.