Artificial intelligence regulation in China: the new draft measures

Just four months after the first legislative measures regarding AI, called "Deep Synthesis Measures on Internet Information Services Administration", the CAC (Cybersecurity Administration of China) introduced the "Deep Synthesis Measures". AI draft". This sudden return to the legislative table appears to have been motivated by a recent surge in international popularity of chat-based generative AI, with new players like Baidu's Ernie Bot and Alibaba's Tongyi Qianwen making inroads into the Chinese market. While the Deep Synthesis Measures focused on audio and video forgery, the AI Draft Measures aim for broader regulation of Generative AI. Importantly, in addition to regulating model outputs, they would also apply to training data and other inputs for generative AI models.

International landscape and social and political implications of the Draft Measures on AI

The AI Draft Measures come at a time of growing international scrutiny over AI, offering an important Chinese perspective on the debate. They outline a regulatory framework in line with the general Chinese approach to regulating data, cybersecurity and online content, with a strong focus on maintaining political and social order. While the CAC proposals reflect many of the substantive considerations found in draft global AI laws and ethical frameworks for responsible AI, the Draft Measures on AI would also require that generative AI reflect core socialist values of China.

Who would be regulated and what would be the regulatory requirements?

The Draft Measures on AI would apply to the research, development and use of generative AI and the provision of services to the public in China. These measures are primarily directed at "Generative AI Service Providers", individuals and organizations that use Generative AI to deliver services. These include chat or text services, image or audio generation, and also include service providers that allow others to generate content through APIs or other means. Before offering a Generative AI service to the public, Generative AI service providers must complete a security assessment and file algorithm records with the CAC.

How would the content of Generative AI be regulated and what would be the data protection obligations?

Providers of generative AI would be obliged to comply with a number of principles listed in the Draft Measures on AI, including the obligation to reflect "core socialist values", to guarantee the accuracy and truthfulness of the outputs, to respect the legal rights and interests and to avoid discrimination based on users' race, national origin or gender. AI-generated outputs should be clearly labeled to signal to the public that they are simulations. In addition to regulating outputs, the AI Draft Measures also focus on inputs to AI models, requiring respect for intellectual property, accuracy, objectivity and diversity of training data, and compliance with data protection requirements.