Abstract

The “Right to Erasure of Personal Information” (right to erasure), which grants individuals the opportunity to delete their personal information from the internet, has become a necessity in the information age. Academic discourse on the right to erasure has centered on the example of search engines as personal information processor and has been set against the backdrop of the internet age. However, the new era of Generative Artificial Intelligence (Generative AI) has given rise to new formidable problems with the protection of the right to erasure, especially in the case of Large Language Models (LLMs) that are trained on large amounts of data containing personal information. This paradigm shift, therefore, raises a host of questions as to how to protect this legal right properly in the age of Generative AI.

Against the backdrop of Generative AI, this paper argues how the Personal Information Protection Law (PIPL) should be applied to govern the protection of the right to erasure in response to the widespread collection and utilization of personal information by LLMs. Through examining unique technical characteristics of LLMs, such as model opacity, operational autonomy, retraining necessity and hallucination possibility, the paper first discusses the major challenges that arise in protecting the right to erasure within LLMs, such as access barrier, deletion dilemma, retraining difficulty, and object absence. The paper then seeks to propose corresponding legal responses to those challenges. To adapt to the evolving context of Generative AI while upholding the right to erasure, it is necessary to introduce third-party supervision that audit databases and filter personal information, to reinterpret the concept of “erasure” in Article 47 of the PIPL, adhere to the principle of purpose limitation throughout the entire personal information processing activities and allow sufficient flexibility for technological advancement. These measures, as the paper shows, would support the technological innovation and sustainable development in the AI sector while providing adequate protection of the right to erasure.

Keywords: Right to Erasure of Personal Information; Large Language Models; Personal Information Protection Law