China's internet watchdog – the Cyberspace Administration of China – and six other PRC authorities yesterday jointly issued the high-anticipated interim measures on the management of generative artificial intelligence services (Measures). The Measures will take effect from 15 August 2023. In-scope service providers will need to act fast.

The Measures follow quickly after China signalled the end of its year-long crackdown on the tech industry, culminating in a high-profile meeting earlier this week between Premier Li Qiang and several Chinese tech giants, including Alibaba, ByteDance and JD.com. Reporting of the pro-business discussion indicates a shift in the Chinese government’s attitudes from reining in entrepreneurship to “increasing their international competitiveness and daring to compete on the global stage”.

The growth of ChatGPT-like AI services has been in the global spotlight, with authorities in the EU, the US, the UK applying tighter regulation to this emerging technology to minimise potential (sometimes unknown) risks.

Do the finalised Measures suggest a more positive move to support generative AI development in China? Below we highlight our key takeaways for tech and other businesses operating in the Middle Kingdom.

AI regulatory landscape in China

With the Measures adding bricks to China’s AI legislative framework, China is clearly seeking a comprehensive AI regulatory regime.

Our regulatory timeline below shows the key rules governing the AI industry:

In-scope generative AI services

The Measures apply to the use of generated AI technology to provide services to the public in the PRC that involve the generation of text, pictures, audio and video content.

  • “Public-facing” element: Chinese authorities have sought to regulate “public-facing” AI services, instead of AI technologies in general or AI-based services intended for internal use. This is a helpful clarification for businesses that intend to utilise ChatGPT-like AI tools for internal business purposes such as improving office working efficiency.

  • No more direct application to developers: Unlike with the earlier draft rules released in April, a developer of generative AI that provides services to the Chinese public is no longer subject to the scope of the finalised Measures – seemingly a response to the industry concern that allocating liability to developers could become a critical constraint on the commercialisation of generative AI in China.

  • Direct and indirect provision: The Measures apply not only to individuals or organisations that directly provide generative AI services, but also those that indirectly provide these services via programming interfaces (or APIs). This lack of a carve-out for indirect provision of services will impact how broader tech ecosystems grow.

Extraterritorial reach

Offshore generative AI providers serving the Chinese public on a cross-border basis will be subject to the new rules – query what the determining factors are to assess the in-scope services “into” the Chinese market.

This extraterritorial application of the Measures may result in legal uncertainty for offshore services providers providing access to their tools to individuals in China. The revisions to the Measures compared to the draft add enforcement teeth by empowering the PRC authorities to shut down network access to offshore services that are deemed to violate PRC law.

OpenAI's ChatGPT and Google's Bard remain inaccessible from China. Despite positive signals in the Measures’ lean towards risk-based control of generative AI tools developed domestically for the mass market, the Measures may continue to pose challenges to wannabe market entrants from overseas.

Classified and graded supervision of generative AI

The Measures instruct that Chinese regulators will take an "inclusive and prudent" attitude towards generative AI services and implement a "classified and graded" regulatory approach.

Although detailed rules are not yet available, multinationals familiar with the development of the EU AI Act will look for a similar regulatory approach to be taken in China – namely, different rules for different risk levels: unacceptable risk, high risk, limited risk, and minimal risk. This would allow international organisations to harmonise operative practices and compliance programmes.

Security assessment and algorithms filing

Aligned with the April draft, the security assessment and algorithm filing obligations remain. Generative AI products and services must undergo a security assessment by the Cyberspace Administration of China, as is the case for online social platforms, and a filing must be conducted following the existing algorithms rules (see our earlier observations on the first batch of filings here). Subject to further details on any variations in the assessments applicable to these AI tools, development plans will need to factor in the increased time and costs of this process.

Looking ahead

Understanding the implications of the Measures and the regulatory trends affecting AI services more broadly will be critical for tech companies and others looking to stay ahead in a competitive and ever-digitalising market.

To answer the question posted at the beginning of this article – compared to the April draft, the finalised Measures have a more supportive tone on promoting generative AI technology, with the Chinese authorities pledging to “promote the healthy development and regulated application of generative AI”. Nonetheless, stringent obligations – including ensuring the authenticity, accuracy, objectivity and diversity of training data; security assessment and algorithm filing processes; gatekeeper liability; and transparency and labelling requirements – seem inevitable in the face of growing national security, data sovereignty and privacy concerns within key stakeholder groups. It also follows that foreign players will be impacted most.

As the name of the Measures suggests, these "interim" rules will – at least for now – govern generative AI services. As part of its ambitious goals to promote the safe development of AI in China, Beijing also has plans to introduce a comprehensive AI Law. As global trade blocs compete for tech supremacy, the timing of this key legislation may be influenced by external factors as much as domestic.

We will be watching, so stay with us and reach out if you have specific questions.