Clifford Chance

The Cyberspace Administration of China (CAC), together with six other PRC regulatory authorities, have jointly published the final version of the 'Provisional Administrative Measures on Generative Artificial Intelligence Services' (GenAI Measures), which will become the first set of regulatory rules applicable to generative artificial intelligence (AI) in China.

Among others, the following key aspects are worth noting:

• application scope – the GenAI Measures apply to any person that utilises generative AI technology to provide services (including, among others, generating text, pictures, audio and video content) to the public in China ('providers'). It should be noted that, other than those directly providing generative AI services, the services indirectly provided in China, e.g., via programming interfaces and APIs, will also be captured by the GenAI Measures. A safe harbour is provided for research on and utilisation of generative AI technologies by enterprises or public institutions, as long as no generative AI services are provided to the public in China;
• foreign investment and regulatory approval – the GenAI Measures reiterate that foreign investors investing in generative AI technology in China shall comply with relevant PRC laws and regulations on foreign investment, and the Providers shall obtain the relevant regulatory approvals before providing generative AI services in China. The CAC retains the power to call for the implementation of technical measures against generative AI services originating overseas that are provided to the public in China and which are not in line with the GenAI Measures and/or other PRC laws and regulations;
• classified and graded supervision – PRC regulators will support the development of generative AI services in a prudent manner, and follow a classified and graded approach in further improving the regulatory framework;
• responsibilities of providers. The providers should be responsible for: (a) the content produced using their services; (b) the legitimacy of the data used to train the underlying generative AI; and (c) the fulfilment of the relevant governance requirements such as a robust complaint handling system. A service agreement shall be put into place between the providers and users of generative AI services;
• filing requirements – any provider providing services that may relate to public opinion or be capable of mobilising or influencing social viewpoints shall complete a CAC security assessment and the regulatory procedures stipulated under the 'Administrative Measures on Algorithm Recommendation for Internet Information Services';
• generated content – the content generated may not contain what is prohibited from circulation in China. Discriminatory content shall be avoided throughout the whole process. The Providers shall also take appropriate measures to improve the accuracy and reliability of the content generated; and
• data used to train AI – the Providers shall, in respect of the data used to train AI, ensure that: (a) data used should originate from legitimate sources and legitimate base models; (b) such data does not infringe on others' intellectual property rights; (c) consent from data subject(s) should be obtained if personal information is included, unless otherwise permitted by laws and regulations; (d) effective measures are taken to improve the authenticity, accuracy, objectivity and diversity of data; and (e) the relevant requirements stipulated under PRC laws and regulations in relation to cyber security, data security and personal information protection are complied with. The Providers are required, upon request from PRC regulators, to explain the sources, volume, types, marking logistics and the algorithm mechanism of data used for AI training, and to provide data and/or technological assistance to the extent necessary.

The GenAI Measures take effect from 15 August 2023.