Research on the Legislative Path of China’s AI Risk Classification and Full-cycle Regulatory——Based on the Experience of the EU AI Act

Abstract

As the world’s first comprehensive AI regulatory legislation, the EU AI Act establishes a dynamic regulatory mechanism covering the whole chain of R&D, deployment, and application through a four-level classification system. In particular, it strictly regulates the implementation of algorithm impact assessment, external entity certification and ex post accountability for high-risk systems, providing an institutional model for the balance between technological innovation and risk prevention and control. China’s current AI legislation is based on the Data Security Law and the Personal Information Protection Law. There are shortcomings in risk classification management, whole chain supervision, and international standard mutual recognition. Learn from the experience of the EU, and enlighten China to establish scenario risk assessment indicators in the fields of health care, finance, etc., support cross sectoral expert committees and dynamic list management, and consider combining the full cycle regulatory logic of the EU with China’s regulatory sandbox mechanism, to improve the whole chain regulatory system from algorithm impact assessment to post responsibility inversion.