China's First Regulation on the Management of Generative AI

Tag：intellectual property, digital economy, telecommunications media entertainment and technology-technology,artificial-intelligence

At present, the new round of technology revolution triggered by generative artificial intelligence (“AI”) is taking the world by storm. Just one month ago, on 14 June 2023, the European Parliament passed the text of the Artificial Intelligence Act (“EU AI Act”) by an overwhelming majority, which places certain obligations upon generative AI systems (e.g. ChatGPT).[1] China is also highly concerned with the regulation of generative AI services. On 13 July 2023, the Cyberspace Administration of China (“CAC”), in conjunction with the national Development and Reform Commission and five other state agencies, jointly issued China’ s first administrative regulation on the Management of Generative AI services - Interim Measures for the Administration of Generative Artificial Intelligence Service (“Chinese AI Measures”). The Chinese AI Measures will come into effect on 15 August 2023, thus leaving only one-month window for those enterprises subject to the Chinese AI Measures to rectify the situation.

The finalized Chinese AI Measures have been substantially revised, after a 3-month public consultation period, compared with the earlier draft the Measures for the Administration of Generative Artificial Intelligence Service ("Draft AI Policy") issued in late April this year (you may refer to our previous analysis of the Draft AI Policy here). Compared with the earlier draft, the finalized Chinese AI Measures give more consideration to the current technological limitations of generative AI. On the other hand, it adopts the AI governance approach as “classified and graded supervision”, thus shaping China’s AI regulation as “top-level general legislation & industry regulation”.

The finalized Chinese AI Measures address a few important topics surrounding generative AI services, including: (1) the scope of this regulation—to whom these AI Measures apply; (2) AI governance—which is based on classified and graded supervision; (3) responsibilities to AI service providers; and (4) protection of personal information.

I. Scope of the Chinese AI Measures

To start with, "Generative AI Technologies", in the Chinese AI Measures, refer to models and related technologies that have the ability to generate content such as text(s), picture(s), audio, and video(s).

Article 2 of the Chinese AI Measures limits the scope of this regulation to those who “provide generative AI services to the public within the territory of the People’s Republic of China (‘PRC’)”, and excludes those (1) who develop and apply generative AI technologies and services for their own “internal use” or “research purposes”—not offered to the public within the PRC, and (2) who provide generative AI services outside the territory of the PRC. Based on the current legislative language, it seems to suggest that if Chinese providers of generative AI services incorporate overseas AI technologies (e.g. ChatGPT) into their own products and then provide such services to the public within the PRC, they are also likely to subject to the Chinese AI Measures.

II. AI Governance: The Idea of “Classified and Graded Supervision”

The Chinese AI Measures seem to follow the same governing principle as with the governance of internet platforms and data compliance in China, where Article 3 and 16 of the Chinese AI Measures propose the governance of AI based on “classified and graded supervision”. However, unlike the EU AI Act, which explicitly sets out a “four-level” risk-based regulatory approach, the Chinese AI Measures have not provided specific criteria for the classification and grading of AI generative services yet.

III. Providers’ Responsibilities

In terms of managing content, providers of generative AI services need to comply with the following principles:

• During the data training process, providers shall only use the data and foundation models that are “legally acquired” and shall not infringe the IP rights of others (Article 7). Also, providers shall improve the “authenticity, accuracy, objectivity and diversity” of training data;
• In terms of AI generated content, providers need to “manage the AI generated content” (Article 14) and take down illegal content or suspend and terminate users’ services, when necessary; and
• Providers are also asked to add “tags” on the AI generated content so as to distinguish such content from human-authored works.

Providers also need to establish a complaint and reporting mechanism (Article 15) to deal with the complaints from the public.

Last but not the least, when providers of generative AI services have the ability to “post/share public opinion” or “mobilize the society”, they are required to undertake “security assessment” and “algorithm filing” (Article 17).

IV. Protection of Personal Information

With regard to the protection of personal information, the Chinese AI Measures prohibit the collection of “unnecessary” personal information and the “illegal” provision of users’ input information to others (Article 11).

As far as the obligation of algorithm transparency is concerned, the Chinese AI Measures only require providers to take “effective measures” to enhance the transparency of generative AI services (Article 4), thus leaving room for providers to fulfill this obligation in more flexible ways. However, compared with the EU AI Act, the Chinese AI Measures again fail to provide specific criteria as to how to judge the transparency of algorithms.

V. Conclusion

The Chinese AI Measures are the first attempt of Chinese government to regulate the generative AI services. Yet a few important issues remain to be clarified, such as the copyrightability of AIGC, the specific criteria for “classified and graded” AI supervision, the definition of the providers’ ability to “support the public opinion or mobilize the public”, and the extent of algorithms disclosure under the algorithm transparency requirements, just to name a few. We expect to see future regulatory guidance as the generative AI landscape continues to evolve.

Scan the QR code to subscribe to "King & Wood Mallesons" for more information