New Generative AI Measures in China

Introduction

Over the past year generative artificial intelligence-based products and services have become widely available – leading to significant public adoption and attention. In turn, governments around the world are continuing to consider how to regulate AI. With its promises of greater efficiencies and transformational capabilities, comes risks relating to data privacy, misinformation, biased or inaccurate answers, and lack of transparency on how decisions affecting people's livelihoos are being made.

Different jurisdictions are diverging in how they regulate AI – with varying views regarding the level of risk that AI poses, what needs to be regulated and restricted, and the role of government regulations versus voluntary commitments.

China has been particularly proactive (by global standards) in developing a regulatory framework of AI – with the aim of balancing fostering innovation versus maintaining control of the technology sector.

On 13 July 2023, the Cyberspace Administration of China (the "CAC"), together with six other PRC government agencies, jointly issued Interim Measures for the Administration of Generative Artificial Intelligence Services (the "Interim Measures") – which came into effect on 15 August 2023. This followed the release of the Measures for the Management of Generative Artificial Intelligence Services (Draft for Comment) on 11 April 2023 by the CAC ("Draft Measures").

The Interim Measures regulate the provision of generative AI services in China and have been substantially amended from the Draft Measures to be more commercial, research and innovation-friendly, as well as recognising possible foreign investment in developing and providing generative AI services in China. Its joint release by seven key central government agencies and ministries (i.e. the CAC, the National Development and Reform Commission, the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration) indicates the importance that the China government is placing on regularing AI.

The Interim Measures will not be the end point for China's efforts in regulating AI. In particular, the State Council has stated that a draft "Artificial Intelligence Law" will be submitted to the Standing Committee of the National People's Congress for review in 2023. While the draft is not yet publicly available, it is expected that it will regulate all forms of AI, including and beyond generative AI.

In the meantime, users and providers of generative AI in China will need to comply with Interim Measures, in addition to related regulations – including regulations on AI recommendation algorithms (i.e. Provisions on the Management of Algorithm Recommendations of Internet Information Services, effective 1 March 2022 – the "Algorithm Measures") and deep fakes (i.e. Administrative Provisions on the Management of Deep Synthesis of Internet Information Services, effective 25 November 2022).

We set out below the key issues that you and your organisation will need to know about the Interim Measures. All Article references are to the relevant Article in the Interim Measures, unless otherwise stated.

What is the scope of the Interim Measures?

The Interim Measures apply to generative AI Service Providers who use generative AI technology to provide services to the public in China. Under Articles 2 and 22:

(a) “Generative AI Technologies” are defined as models and related technologies that can generate text, pictures, audio, video, and other content.

(b) “Generative AI Services” are defined as services that are offered to the public and that can generate text, pictures, audio, video, and other content by use of generative AI technologies (references in this article to "services" are to such Generative AI Services).

(c) “Generative AI Service Provider” refers to an entity or individual that uses generative AI technologies to provide Generative AI Services, including providers of the underlying generative AI technologies and any person who provides applications on top of such technologies via APIs.

(d) "Generative AI Service User" refers to an entity or individual that uses contents generated from the Generative AI Services.

How "public" is defined is a key point in the Interim Measures. Article 2 states that the Interim Measures apply to services that are directed to the domestic public (e.g. Baidu's ERNIE bot and Alibaba's Tongyi Qianwen) and clearly excludes industry associations, enterprises, education and research institutions, public cultural bodies and related professional bodies that research, develop and/or use generative AI technology without providing public-facing services – this is a key difference from the Draft Measures. However, the Interim Measures are not clear on whether "domestic public" means consumers only (i.e. B2C) or includes businesses as well (i.e. B2B) – noting that certain parts of the Interim Measures appear to contemplate B2C generative AI use.

Article 20 states that Service Providers that provide services to the public in China from outside China may still be subject to the Interim Measures (the Draft Measures were silent on extraterritoriality) – and that if those services do not comply with the Interim Measures, then they will be subject to technical measures that CAC and other PRC regulators may implement (in practice, this likely means that non-compliant services will have access within China restricted or blocked).

Content and data management

Generative AI Service Providers need to ensure that all content created and services provided comply with requirements pursuant to the following five principles:

(a) Uphold the core socialist values and not generate content that may harm national security or that is forbidden by laws or regulations.

(b) Adopt effective measures to prevent the content from discriminating on any basis.

(c) Respect Intellectual Property rights and commercial ethics.

(d) Respect the lawful rights of others, and not infringe on their well-being and interests.

(e) Undertake strict and effective measures to increase the transparency, accuracy and reliability of their services.

While the above requirements are substantial, they also reflect obligations seen in draft and passed AI laws and regulations around the world.

Notably absent from the Interim Measures is the proposal that was in the Draft Measures that the Service Provider must be "capable of ensuring" the accuracy of the content generated – there were significant public objections to this and at a practical level it was unclear whether Service Providers could comply with such a threshold. The Interim Measures implicitly recognised this feedback and effectively replaced such proposal with the paragraph (e) requirement above.

Service Providers' obligations

The Interim Measures generally set out that Services Providers are required to cooperate with regulators' supervision and inspection, including providing necessary support and assistance.

Specific obligations under the Interim Measures for generative AI Service Providers include:

• Service Agreement and identity information. A service agreement must be executed between the Service Provider and the users who register for its services. Note that the Draft Measures requirement for real identity information from users has been removed, though it remains to be seen how this will apply in practice given other real-name requirements in China.
• Anti-addiction measures. Service Providers must implement effective measures to prevent minor users from becoming addicted to such services. This is consistent with CAC's efforts in other technology areas, in particular video games.
• Labelling and normal uses. Service Providers who offer "deep synthesis" (or "deep fake") services must label any visuals or videos to alert the public about their non-human authored nature. Service Providers must also ensure that their service is safe, stable and continuous.
• Illegal content and services. Service Providers must immediately suspend and terminate the generation or transmission of any illegal content and services. In addition, if any users are found to be using the services for illegal activities, Service Providers must take actions to suspend services, issue warnings and report the case to relevant authorities.
• Pre-training and optimization training. Service Providers must only use data that has been "legally acquired" for model training. Article 19 requires Service Providers to provide information on the training data's sources, models, types, tagging rules and algorithm mechanisms to relevant government departments (though noting that the Interim Measures do not outline what supporting documents are required). The Draft Measures included a requirement to take measures within three months through model optimization training to prevent illegal content from being generated again – this has been removed from the Interim Measures.
• Personal information collection. Service Providers must not collect unnecessary personal information, store input information or usage records in a way that can identify the personal details of users, or provide such information or records to third parties.
• Compliant channels. Service Providers must establish a mechanism for receiving and handling user complaints. In addition, they must also respond to any user's request for reviewing, copying, correcting, supplementing and deleting their personal information in a timely manner.
• Security assessment and algorithm filing of certain services. If any service possesses public opinion attributes or have capacity for "social mobilisation", the Service Provider must conduct a security assessment, and comply with the filing/alteration/cancellation-related requirements set out in the Algorithm Measures. In particular - the security assessment can be conducted either by the Service Provider or through a third party, and must be filed with the local branches of the CAC and China's Public Security Bureau. Such filing of an algorithm must be completed within ten working days of providing relevant services.

In addition, if foreign investors intend to invest in the generative AI services in China, such investment must also comply with the relevant foreign investment related laws and regulations.

Impact and outlook

The Interim Measures relax some of the requirements set out in the Draft Measures, reflecting the China government's intention to encourage development of generative AI service – while at the same time setting out obligations regarding content management, protection and security of personal data, and transparency of generative AI within China. The Interim Measures also represent a new addition to China's regulatory framework for data.

There are still some questions to be answered. Notably, the Interim Measures also state that there may be further classified and graded regulations applicable to generative AI services. We await with interest what such regulations may be. In addition, some of the aspects within the Interim Measures will likely be developed – such as copyright status/protection of any materials arising from generative AI, how and the extent to which algorithms are to be disclosed, and how phrases such as “support the public opinion or mobilize the public” will be interpreted.

With significant investments continuing in the AI sector and with AI being a key tenet of China's "Made in 2025" industrial plan, we expect further generative AI (and, more generally, all AI) regulations and regulatory guidance to be forthcoming. We will continue to closely monitor developments while working with clients in a variety of industries and sectors in this evolving space.

With thanks to Derek Wang, Senior Associate; Chloe Xiang, Associate; Shashtika Sundar, Trainee Lawyer; Zoey Li, Intern for their contributions to this article.

Disclaimer: Ashurst Australia (ABN: 75 304 286 095) is a general partnership constituted under the laws of the Australian Capital Territory and is part of the Ashurst Group. The Shanghai Representative Office is a representative office established by Ashurst Australia in China and is not licensed to practise PRC law. If you require a formal legal opinion on PRC law or any other specifically PRC law-related assistance, we can issue a formal PRC legal opinion or provide specific legal services in relation to PRC law through Ashurst's Joint Operating Office with Guantao Law Firm.