PrivAct: 通过多智能体偏好训练内化上下文隐私保护
PrivAct: Internalizing Contextual Privacy Preservation via Multi-Agent Preference Training
AI总结:
PrivAct通过多智能体偏好训练,内化上下文隐私保护,提升模型生成行为的隐私合规性,减少信息泄露并保持实用性。
摘要:
大型语言模型(LLM)代理越来越多地被部署在涉及敏感、依赖上下文的信息个性化任务中,其中由于上下文隐私的隐含性,代理的行为可能导致隐私违规。现有方法依赖于外部的推理时间干预,这些干预是脆弱的、场景特定的,并且可能扩大隐私攻击面。我们提出了PrivAct,一种上下文隐私感知的多代理学习框架,该框架将上下文隐私保护直接内化到模型的生成行为中,以实现符合隐私的代理行为。通过将隐私偏好嵌入到每个代理中,PrivAct在增强系统整体上下文完整性的同时,实现了更有利于隐私有用性权衡的改进。在多个LLM基础架构和基准测试中的实验显示,上下文隐私保护的一致改进,泄漏率减少高达12.32%,同时保持相当的有用性,以及在多样化的多代理拓扑中实现零样本泛化和鲁棒性。代码可在https://github.com/chengyh23/PrivAct上获得。
英文摘要:
Large language model (LLM) agents are increasingly deployed in personalized tasks involving sensitive, context-dependent information, where privacy violations may arise in agents' action due to the implicitness of contextual privacy. Existing approaches rely on external, inference-time interventions which are brittle, scenario-specific, and may expand the privacy attack surface. We propose PrivAct, a contextual privacy-aware multi-agent learning framework that internalizes contextual privacy preservation directly into models' generation behavior for privacy-compliant agentic actions. By embedding privacy preferences into each agent, PrivAct enhances system-wide contextual integrity while achieving a more favorable privacy-helpfulness tradeoff. Experiments across multiple LLM backbones and benchmarks demonstrate consistent improvements in contextual privacy preservation, reducing leakage rates by up to 12.32% while maintaining comparable helpfulness, as well as zero-shot generalization and robustness across diverse multi-agent topologies. Code is available at https://github.com/chengyh23/PrivAct.
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。