A Quick Glance at China’s First Regulatory Endeavour on Generative AI
Raymond Wang and Yihan Zang of Shihui Partners discuss China's efforts to regulate generative AI.
China has recently joined the global regulatory race in response to the explosive growth of generative artificial intelligence (AI) technology. In July 2023, the Cyberspace Administration of China (CAC), in consultation with six other regulators, issued the Interim Measures on the Management of Generative Artificial Intelligence Services (the “Measures”), which came into force on 15 August. Compared to the two regulations on algorithms and deep synthesis that were released in 2022, the Measures do not bring substantial new obligations, with the vast majority of the rules being on the application of existing legal requirements in the realm of generative AI technology (or large language models). Notably, the Law on the Progress of Science and Technology is explicitly referred to as an upper-level legal source of the Measures, signalling that the government’s overall policy goal is to promote the development and application of generative AI.
Scope of Application
The Measures primarily regulate “generative artificial intelligence services” – ie, the provision of generated text, images, audio, video and other content through the use of generative AI technology. The emphasis is on the oversight of “services” as opposed to the “technology”. This approach echoes the methodology of China’s existing regulations on algorithms, and leaves room for technological experimentation and growth.
"The government’s overall policy goal is to promote the development and application of generative AI."
The Measures only apply to those who provide generative AI services to the “public” in the territory of China. Therefore, the internal research, development and application of generative AI technology carried out by business undertakings do not fall within the purview of the Measures; the provision of generative AI services within China to international clients is also excluded from their coverage.
Regulatory Approach
The Measures regulate generative AI services with the aim of advancing the following policy goals:
- safeguarding national security and interests, maintaining social stability and upholding good moral values;
- preventing discrimination;
- preserving intellectual property, commercial ethics and trade secrets;
- promoting antitrust measures and preventing unfair competition;
- protecting the legal rights of others; and
- enhancing the transparency of the services, and the accuracy and reliability of the generated content.
In light of the technical features of generative AI, the Measures set forth requirements that span all the way from the model's development to its application. This lifecycle-based oversight can be broadly divided into algorithm compliance and service compliance. Algorithm compliance mainly involves supervision of the legality and legitimacy of training data and foundational models. It covers such obligations as respecting intellectual property rights in the training data, ensuring the legality of personal information used in the training, maintaining a certain level of accuracy and quality of the training data, and fulfilling manual labelling obligations.
"Lifecycle-based oversight can be broadly divided into algorithm compliance and service compliance."
Service compliance is manifested in various facets during the provision of the services, such as the detection and disposal of unlawful content, the collection and protection of input data, the prevention of addiction by users, and the handling of complaints and reports. It also involves constraints on the use of the services for unlawful purposes.
Similar to the regulations on algorithm recommendation and deep synthesis, the Measures also require companies to file algorithms and conduct safety assessments under certain circumstances, especially when the services come with attributes of public opinion or social mobilisation. However, when developers of generative AI models collaborate with companies in specialised fields to optimise, train and provide services, the specific duty owners for algorithm filing and safety assessment still need further clarification in practice.
Looking Ahead
The Measures represent China’s first attempt to regulate generative AI, and are purposefully reactive, transitional and structural. Some rules, such as classified monitoring, are only briefly touched upon in the Measures and require further elaboration. The State Council of China has included a Draft Artificial Intelligence Law in its deliberations for the 2023 legislative work plan for the National People's Congress Standing Committee. In the future, generative AI in China will likely be governed by a holistic legal framework based on an umbrella legislative law, much like what is being contemplated in the EU.
Shihui Partners
