China and AI Regulation: The Most Comprehensive Framework in the World and its Impact on Global Companies
China has built the world's most detailed and layered AI regulatory framework. With three main regulations already in effect and at least two more in preparation, the country requires registration of foundational models, pre-launch security assessments, and control of AI-generated content. Companies with operations or users in China need to map compliance across multiple fronts simultaneously.
While the AI regulatory debate in the United States remains fragmented between states and the federal Congress, and Europe implements the AI Act with staggered deadlines through 2027, China has constructed the world's most detailed and operationally specific AI regulatory framework, with rules already in place for specific use cases.
The Layered Structure
Deep Synthesis Provisions (in effect since 10 January 2023): regulate the creation of synthetic content, including deepfake videos, voice cloning, and AI-generated texts. They require that synthetic content be clearly labelled, that service providers verify the real identity of users, and obtain consent before using the likeness of real people. Creators of deepfakes without consent are subject to administrative and civil penalties.
Generative AI Measures (in effect since 15 August 2023): apply to any generative AI service offered to the public in China, including chatbots, image generators, and text models. They require pre-launch security assessments for services with relevant "public opinion" or social mobilisation capacity. Foundational models need to be registered with the competent authority (Cyberspace Administration of China, CAC).
Algorithm Recommendation Measures (in effect since 1 March 2022): impose transparency on content recommendation systems, prohibit the use of algorithms to create "addiction dependency", and require an opt-out mechanism.
Impact on Foreign Companies
Any company offering AI products to users in China, regardless of where it is based, is subject to these regulations. The requirement for registration of foundational models is particularly relevant: the CAC maintains a database of approved models, and operating with an unregistered model could result in service blockage.
For companies with local operations in China, the requirements for storing data created by Chinese users on servers within the territory (Data Security Law and Personal Information Protection Law, both in effect since 2021) add to the AI requirements, creating a regulatory stack that typically requires dedicated infrastructure within the country.
The Tension with AI Development
The Chinese regulatory framework presents a structural tension: while imposing strict requirements on models offered to the public, the government actively encourages the development of AI in strategic sectors such as manufacturing, healthcare, and finance. The logic is one of social control over AI output for consumers, combined with support for the development of national technological capability.
For C-level executives of companies with expansion plans in China or significant supply chains in the country, the question is not whether the regulations apply. It is whether the organisation has the legal and technical resources to navigate a framework that changes more frequently than any other global jurisdiction.