Generative AI Regulations Officially Released in China

View PDF

Authored by: Edwarde Webre and Jenny Jin

On 11 April 2023, the Cyberspace Administration of China (“CAC”) published the draft Administrative Measures for Generative Artificial Intelligence Services for public comment. On 10 July 2023, after months of deliberation and extensive revision, the CAC and several ministry-level departments jointly released the final version of the regulations, the Interim Measures for Administration of Generative Artificial Intelligence Services (the “Interim Measures”), which will take effect on 15 August 2023. The Interim Measures generally adopt a friendlier approach than the draft measures to the regulation of the AI sector, attempting to balance the development and governance of the sector. The Interim Measures are narrower in scope than the draft measures and impose fewer obligations on service providers and developers.

Scope of Application

The Interim Measures apply to the provision of generative AI services to the public in China that utilize generative AI technology to generate text, pictures, audio/video and other such content. They are not applicable to research, development and application of the generative AI technology that do not involve the provision of generative AI services to the Chinese public. Generative AI technology is defined as models and relevant technology capable of generating texts, pictures, audio, video and other type of content.

The Interim Measures shall also apply to the provision of generative AI services to the Chinese public from abroad. Such services must comply with PRC laws and regulations including the Interim Measures. In the event of non-compliance, the CAC may notify the relevant organizations (which could include the in-charge PRC authorities and the PRC local partners of the foreign AI service providers) to take technical and other measures to deal with the non-compliance. The Interim Measures do not elaborate how this may be handled, but blocking access to the services or sanctioning the services providers may be the primary options. Consequently, foreign companies even without operations in China may also be targeted if they provide generative AI services to China or such services are accessible by the Chinese public.

Using the generative AI technology to engage in news publication, film and TV production, cultural creation, etc. may be subject to the industry-specific regulations. Foreign investment in the generative AI service sector should also comply with the PRC foreign investment laws and regulations.

Government’s Support and Supervision

The PRC government will adopt various measures to promote the development and application of the generative AI technology, such as constructing the generative AI infrastructure and a public platform for data resource training and sharing computing resources. To strike the balance between regulating and promoting the AI sector, the PRC government will adopt an inclusive and prudent approach to establish a classified and graded regulatory regime. It is not clear what elements (such as risks, industries, etc.) will be factored in to the classification and gradation of the governance of the AI services, but the government’s plans and strategies for the future regulation of the AI sector have been indicated.

General Obligations for Provision and Use of Generative AI Services

The providers and/or users of the generative AI services are required to uphold the socialism core values, avoid fake and harmful information, respect third-parties IP rights, trade secrets, privacy and other personal rights, and not use such services in any monopoly or unfair competition activities. The providers should also take effective measures to prevent discrimination when designing algorithms, selecting training data, creating and optimizing models and providing services, and improve the transparency of the services and the accuracy and reliability of the contents.

Specific Obligations for Providers of Generative AI Services

The providers of the generative AI services should:-

• Conduct data training in a legal manner. The source of data and the foundation models should be legal and free from IP infringement and with individual consents unless otherwise exempted. Effective measures should be taken to improve the quality, authenticity, accuracy, objectivity and diversity of the training data;
• Annotate data properly. Annotation rules should be made, annotation quality assessment should be conducted and training should be provided to the annotators;
• Assume responsibility as a producer of the internet information content and a personal data processer and fulfil the relevant obligations. The providers should sign contracts with their users to set out the parties’ rights and obligations. The Interim Measures do not specify the contents of the contracts. It remains to be seen if and to what extent the providers may shift their responsibilities to the users by way of contract;
• Guide users to use their AI services legally and take effective measures to prevent minors’ overreliance upon or addiction to their AI services;
• Protect users’ input information and use records and not collect unnecessary personal information. The user’s input information or use records may not be illegally stored or provided to others;
• Label the AI generated content in accordance with the PRC Administrative Provisions on Deep Synthesis of Internet-based Information Services (“Deep Synthesis Provisions”);
• Provide safe, stable and continuous services to ensure users’ normal use;
• When discovering any illegal content or illegal use of their AI services, take measures to cease the generation or transmission of the illegal content and remove the content from the services, implement optimizing model training. The event must be reported to the authorities. The service provider may need to suspend or terminate services to the violators;
• Establish a complaint mechanism to deal with public complaints;
• If its AI services are capable of influencing public opinion or causing social mobilization, the AI service provider must carry out a security assessment and algorithm filing in accordance with the Administrative Provisions on Recommendation of Internet-based Information Service Algorithms and other related regulations. It is notable that under the Deep Synthesis Provisions, certain deep synthesis services (such as generation or edition of human face, voice or other biological identification information) which are not necessarily capable of influencing public opinion or causing social mobilization are also subject to security assessment. The generative AI services belonging to this type of deep synthesis services may also require a security assessment.

Liabilities for Violations

Violations of the Interim Measures will be punished according to the PRC Cyber Security Law, the PRC Data Security Law, the PRC Personal Information Protection Law and other relevant laws and regulations. In the absence of relevant provisions in the foregoing laws and regulations, violators may be given a warning or criticised and/or be ordered to make rectification within a prescribed period. In case of failure to obey the rectification order or under serious circumstances, the violators may be ordered to suspend provision of relevant services.

Conclusion

While the Interim Measures are general and broad pending further fine-tuning and clarification, they lay out the general direction of the regulation of the generative AI service sector. On 6 June 2023, the PRC State Council released its legislation plan for 2023 indicating its intent to submit a draft PRC Artificial Intelligence Law for review by the country’s top legislative body the PRC National People’s Congress. More comprehensive legislation is sure to be coming. It’s time for companies engaged in the AI business to review their operations and beef up their compliance efforts.

Key Contacts

Edwarde Webre

Consultant | Corporate Commercial

Email or call +852 2825 9730

Jenny Jin

Legal Counsel | Corporate Commercial

Email or call +86 21 6340 3588