The emergence of Agentic AI, such as ChatGPT and Gemini, which possesses active reasoning capabilities, is fundamentally transforming how digital services are utilized. Hyper-personalization services based on Agentic AI provide optimized benefits by i...
The emergence of Agentic AI, such as ChatGPT and Gemini, which possesses active reasoning capabilities, is fundamentally transforming how digital services are utilized. Hyper-personalization services based on Agentic AI provide optimized benefits by integrating heterogeneous data distributed across multiple controllers in real time, evolving into a more sophisticated form of service that surpasses traditional, one-dimensional customization. The realization of such services necessitates a process of collecting and synthetically analyzing personal financial and lifestyle data scattered across various platforms without boundaries. However, the autonomous data integration methods of Agentic AI structurally collide with the strict dichotomous framework of "entrustment" and "third-party provision" under the current Personal Information Protection Act. If the transfer of data to Agentic AI is interpreted as a simple entrustment relationship, it becomes difficult to control the AI provider—as a trustee—diverting data for independent purposes such as model training, which creates a regulatory vacuum where liability is unfairly shifted to lower-level service providers. Conversely, if this is viewed as third-party provision between independent controllers, individual consent must be obtained at every stage of vast data integration, leading to practical bottlenecks and extreme consent fatigue for data subjects. Furthermore, as these services are predominantly led by global companies headquartered overseas, there are inherent sovereign risks, such as potential data leaks due to information disclosure orders from foreign judicial authorities and the loss of domestic judicial control. Accordingly, this study analyzes the regulatory blind spots created by the current legal system and proposes a new liability-based legal framework that ensures the effective remedy of data subjects' rights while fostering industrial innovation. As specific legislative improvements, this study first suggests the introduction of a "Joint Personal Information Controller" system, drawing on the concept of Joint Controllers under the EU GDPR, to enforce joint and several liability among entities within complex processing environments. Second, moving beyond the structural constraints of the current MyData authorization system, this study proposes the creation of a "Specialized Agentic AI Data Recipient" status, which allows for the legal receipt of data across all industrial sectors upon demonstrating high levels of internal control and proactive rights protection systems. Third, the study establishes post-facto purification liability for companies by codifying "Algorithmic Disgorgement" orders, which mandate the disposal of intelligent models derived from illegally processed information. Finally, the study defines measures to safeguard data sovereignty by imposing a judicial defense obligation on domestic representatives against unjust compulsory execution by foreign judicial authorities within global AI processing structures. This study holds academic significance by re-examining the risks of privacy infringement and liability evasion in the Agentic AI environment through the lens of substantive regulation and structural attribution of liability to controllers, rather than the passive defense logic of data subjects. It is expected that the alternatives proposed in this paper will serve as a legislative cornerstone and a normative foundation for future amendments to personal information protection laws and the establishment of new regulatory governance for the era of Agentic AI.