Privacy threats and solutions in the zero-click era of agentic AI

Lee Seong-yeob

The author is a professor at Korea University’s Graduate School of Management of Technology and president of the Korean Society of Information and Communications Law.

For decades, we have navigated the vast ocean of the internet through clicks. Searching for information, adding items to a shopping cart and pressing the payment button have all represented not only user decisions but also expressions of consent. That model is now rapidly disappearing. The era of zero-click computing, in which AI searches for information, makes decisions and carries out tasks without requiring user input, is arriving much faster than many expected.

At the center of this shift is agentic AI, a new generation of AI systems designed not merely to answer questions but to understand users' goals, develop multistep plans and carry them out autonomously on their behalf. Conventional generative AI answers questions, but agentic AI executes tasks. Amazon's Buy for Me service, for example, allows users to request recommendations for products such as running shoes. AI searches not only Amazon but also external online stores, presents suitable options and, once the user makes a selection, fills in shipping and payment information on the seller's website before placing the order. Naver's AI shopping agent is evolving in a similar direction by recommending products before users even begin searching.

While this technology reduces the burden of repetitive clicks, it also creates unprecedented privacy risks that are difficult for individuals to control. Those risks can be understood in three broad categories.

The first is the disclosure of personal information without meaningful consent. Modern privacy protection begins with the principle that individuals decide who receives their data and for what purpose. In a zero-click environment, that premise weakens. As AI agents visit multiple websites, agree to terms of service and provide personal information on behalf of users, individuals gradually lose practical control over how their data is shared.

The second concern is AI's ability to remember and retain enormous amounts of information. Agentic AI seeks to understand users by recording what they search for, what they buy and even the relationships they maintain. Over time, a detailed record of an individual's life accumulates. Principles such as data minimization and the right to be forgotten become increasingly difficult to uphold because AI systems derive much of their value from preserving long-term contextual memory about individual users.

The third issue is the emergence of a surveillance society through data aggregation. Agentic AI integrates email, messaging services, social media and location data into a unified system. Each dataset may reveal little on its own, but together they enable highly detailed profiling capable of reconstructing nearly every aspect of a person's life. The concentration of such power in large platform companies raises concerns that extend beyond privacy to democratic governance and individual freedom.

These developments suggest that the traditional consent-based model is no longer sufficient. Requiring users to approve every step would eliminate the convenience of zero-click services, while granting blanket authorization would effectively surrender meaningful control. Consent risks becoming little more than a procedural formality that shields companies from liability.

The priority, therefore, should shift from consent to system design. Privacy by Design, a framework that incorporates privacy safeguards from the earliest stages of system development rather than adding them afterward, should be embedded in AI services. It should be accompanied by mechanisms that ensure traceability throughout data processing and clearly assign responsibility for AI actions. Government oversight should likewise move beyond punishing individual data breaches after the fact and instead focus on comprehensive risk management, including privacy impact assessments and reviews of algorithmic transparency. At the same time, regulatory standards must be carefully calibrated so that excessive intervention does not undermine innovation.

More fundamentally, AI's ability to collect and process vast quantities of data through opaque decision-making systems demands a rethinking of existing privacy frameworks. This challenge cannot be addressed through nonbinding guidelines alone. Even the world's leading economies, despite regulatory initiatives such as the European Union's AI Act, have yet to develop a comprehensive solution to these privacy challenges. Korea should therefore consider enacting a special law tailored to personal information protection in the AI era.

Protecting personal information is no longer simply the responsibility of individual users. It has become a core design principle for AI systems and a central task of national governance. The convenience of zero-click services may be inevitable, but the disappearance of clicks must not also mean the disappearance of user control. Ensuring that control remains intact is, to a significant degree, the responsibility of the state.