these AI applications operate on smartphones, they may request access to device-level data or functions such as contacts, location information, or stored photos. While users can technically control each permission individually, understanding and managing a large number of settings in a comprehensive and accurate manner is not an easy task. As AI systems evolve beyond “generative” functions and increasingly operate as “agents” that act on behalf of users, the risks to personal data protection are likely to grow significantly. When multiple agents exchange data—such as an AI agent on a user’s smartphone communicating with an airline’s AI agent—the flow of information becomes far more difficult to track than it is today. Even if the user issues instructions and intermittently monitors the process, the detailed steps required to carry out those instructions are typically executed autonomously by the agent. As a result, it becomes harder to determine who has access to personal data, how long transmitted data is retained, and whether it is being properly managed. This increases the risk of inadequate protection or intentional misuse. The growing number of data transfers also heightens the risk of security breaches, and delegating account access to agents raises the possibility that accounts may be manipulated without the data subject’s awareness. In this context, policy measures such as limiting data transfers to the minimum necessary and ensuring the deletion of data once its purpose has been fulfilled become even more critical, in line with core personal data protection principles. In addition, AI providers, as data controllers, should be subject to stronger obligations to 84 85