Privacy preference control for agent systems
Privacy Preference Controls for Agentic Systems
The paper responds to the trend of the web shifting from user-driven to agent behavior and proposes an actionable taxonomy of privacy preferences for agent systems. The taxonomy was developed based on relevant laws, empirical research on existing privacy tools, and an audit of 30 mainstream AI assistant and agent platforms. The authors also developed 8 open-source prototypes simulating real agent applications, demonstrating how the main aspects of the taxonomy can be structurally integrated into agent protocols such as the Model Context Protocol and the Agent2Agent protocol. The research aims to lay the foundation for ensuring that end users' privacy and autonomy are not lost in the coming agent world.
Related events and news
1 events · 1 news mentions
News that mentions this paper
评论文章主张为AI智能体设数据最小化规则
- Tech Policy Presscites the paper's id or link评论文章主张 AI 智能体须建立数据最小化规则
It is a necessary foundation for what researchers call Privacy Preference Controls for Agentic Systems.