Privacy Control Concerns Become Reality Amid Craze for 'Pickle,' the AI That Acts on Its Own

As App Connections Grow, So Does the Use of Personal Information
From Inferring Family Relationships to Making Unrequested Calls
More Convenient AI Assistants... But Controlling Their Permissions Remains a Challenge

"How did you even know about my family relationships when I never told you?"

A reconstruction of a chat in which AI agent "Pickle" identified family relationships that were not explicitly stated.

Office worker A was surprised to find that the AI agent Pickle, which has attracted attention as Korea's version of "Muse," had figured out their children's names and birthdays. Pickle inferred the family relationships without being told by combining security alerts for the children's accounts in the Google mailbox A had linked, Netflix profiles, and contacts.

Office worker B learned that Pickle had called several nearby restaurants without permission. Pickle explained why it had done so, saying, "I acted on my own judgment while preparing the morning briefing." B subsequently instructed Pickle to obtain approval before making calls or sending messages or emails.

As of November 9, concerns about privacy and excessive autonomy are becoming reality as AI agents evolve to the point where they can handle everyday tasks on people's behalf from start to finish. The way these agents carry out tasks using information from connected external apps and services has led to frequent cases in which they gather information scattered across multiple services to infer users' personal details, or take actions on their own that users never explicitly requested.

Pickle's strong compatibility with domestic services, in addition to overseas Big Tech services such as Google, is considered one of its key strengths. Pickle says it can connect with apps and services in sectors including telecommunications, delivery, insurance, and aviation, such as KakaoTalk, Government24, and Coupang. The company says authentication information such as passwords is encrypted and cannot be viewed directly by the AI, but that email contents and schedules retrieved through these services may be incorporated into conversations and memory.

Meanwhile, the effective date of the privacy policy posted on Pickle's official website is December 8, 2026. The service is already operating using users' accounts and personal information, but its published privacy policy is set to take effect a month later.

Experts say the institutional framework supporting AI agents' autonomous task execution needs to be updated, and the scope of their authority clearly defined. Park Jinho, a professor in the Department of Computer Science and Artificial Intelligence at Dongguk University, said, "In Korea, an institutional framework has yet to be established for AI agents' automatic ordering and payment or the scope of their use of personal information," adding, "Users also need to make clear decisions about what authority to grant an agent." Choi Daeseon, a professor in the School of Software at Soongsil University, said, "The focus of personal information protection is shifting from 'data' to 'actions.'" He added, "When agents access external systems, they should be required to go through a designated proxy server, and non-deletable logs should be kept of what information they accessed and transmitted."

This content was produced with the assistance of AI translation services.