测评7个手机智能体：点一杯奶茶，意味着交付40%高敏权限

Core Insights - The rise of intelligent agents signifies a shift from machines merely chatting to machines performing tasks on behalf of users, with a notable focus on local deployment rather than cloud-based solutions [1][2] - The increasing integration of intelligent agents into personal devices raises concerns about privacy and the extent of permissions required for their operation [2][19] Group 1: Intelligent Agent Development - Intelligent agents like Doubao are being embedded in smartphones, allowing them to perform various tasks such as ordering food and playing games [3][4] - A significant number of permissions are required for these agents, with many devices exceeding 100 system permissions, which is notably higher than typical applications [5][6] Group 2: Permission and Privacy Concerns - Approximately 40% of the permissions requested by mobile intelligent agents are classified as high-sensitivity, including access to precise location and reading messages [7][10] - The use of accessibility permissions allows intelligent agents to bypass standard security measures, raising concerns about potential misuse and privacy violations [10][12] Group 3: Security Measures and User Awareness - Companies like Doubao and OPPO have begun to publish AI privacy and security white papers, focusing on user awareness and control over sensitive operations [24][25] - The need for transparent usage guidelines and the importance of logging AI operations have been emphasized to ensure accountability and traceability [25][29] Group 4: Data Handling and Cloud Risks - The process of data transmission to the cloud remains controversial, with concerns about sensitive information being exposed during this process [29][30] - Current intelligent agents struggle with effectively identifying and protecting sensitive information, highlighting a gap in privacy protection capabilities [30][32]