context: The PRC has built its AI regulatory framework through a series of targeted, application-specific rules—algorithmic recommendations (2022), deepfakes and synthetic content (2022), generative AI (2023), facial recognition (2023)—rather than through a single comprehensive statute. Against the backdrop of having established landmark legislation in adjacent domains—Cybersecurity Law (2017), Data Security Law (2021) and Personal Information Protection Law (2021)—there has been sustained expectation that a comprehensive AI law would follow as the natural capstone. This article argues conditions are not yet ripe for unified AI legislation, and the technology remains too fast-moving and uncertain to codify comprehensively. This preserves both regulatory flexibility and maximum state discretion—keeping the framework adaptive at a moment of intense global AI competition.
Zhi Zhenfeng 支振锋 Chinese Academy of Social Sciences Institute of Law Xi Jinping Rule of Law Thought Research Office director, published an article in Legal Daily summarising the PRC's experience in AI legislation and setting out priorities for the next stage
- why high-quality AI legislation is needed
- security: guard against risks including ethical failures, data security breaches, algorithmic discrimination and deepfake misuse
- development: provide a stable, transparent legal environment for the flow of data, algorithms and computing power
- public welfare: regulate facial recognition misuse, manipulative personalised recommendations and misleading generated content
- global governance: build a sound legal framework to contribute to the PRC's approach to global AI governance
- the PRC's AI legislative experience: the 'small, fast, flexible' approach
- civil, criminal, cybersecurity, data security and personal information protection laws provide the overarching framework
- targeted departmental regulations address specific issues—algorithmic recommendations, generative AI, facial recognition, deepfakes—with speed and precision
- strategy: 'urgent needs first, mature issues first, dynamic refinement'—no pursuit of a comprehensive unified law
- multi-stakeholder governance embedded in legislative design: government regulation, corporate responsibility, industry self-discipline and public participation
- priorities for the next stage
- explicitly opposed: conditions are not yet ripe for a national unified AI law
- use existing laws to hold the security baseline; be cautious about new legislation
- place innovation and development support at the forefront: protect R&D investment, facilitate data factor flows, implement regulatory sandboxes and error-tolerance mechanisms
- strengthen personal information protection; prevent algorithmic discrimination; clarify liability boundaries for developers, providers and users
- advance global AI governance: implement the Global AI Governance Initiative; oppose technology barriers and tech hegemony