Following our webinar on AI legal developments in Asia, we’ve published a series of three articles looking closely at the different countries. This third piece looks at China.
(The first article covers Singapore, Indonesia and Malaysia, while the second studies Vietnam, Thailand and the Philippines.)
AI development has been included in China’s national strategy since 2016. Today, the country has a plan to set up an integrated big-data system with eight computing hubs placed around the country. By the end of 2023, the AI industry in China is estimated to have reached a size of RMB 578 billion (USD 81 billion).
“According to figures from China’s Academy for Information and Computing Technology, computing power in China accounted for some 33% of total global output by 2022. The country was second in terms of computing power output in 2023, reaching 230 EFLOPS. This shows China’s support for and efforts in AI,” says Landy Jiang, Managing Partner at Lusheng Law Firm (Rouse’s strategic partner in China).
Due to network-access and account-registration rules, most foreign AI tools – such as ChatGPT and Microsoft Copilot – are not directly available in mainland China. A number of local AI applications have been developed instead. By the end of March 2024, the number of Large Language Models (LLMs) officially recorded in China stood at 117.
There is currently no AI Act in China, although more than 20 policy documents related to AI have been published. Many of these documents point to the same basic principles for AI development as seen in other jurisdictions. These include provisions for respecting fairness, privacy, safety, open collaboration and agile governance.
“China has been encouraging the development of AI and a friendly environment for the technology, so governance is not that strict yet,” says Jiang. “The country currently uses a filing system to launch generative AI models to the market. This is more lenient than an approval system.”
Chinese courts rule on copyright protection
As in other markets, China is also dealing with issues around copyright protection for AI-generated content.
“In principle, content created by a machine needs to clearly demonstrate the involvement of human intelligence, efforts and control in order to be copyrighted,” explains Jiang. “The other big question is around who should enjoy copyright protection for AI-generated content: users, AI model developers, and/or AI service providers?”
In 2019, the Chinese court recognised that AI content generated by Tencent's Dreamwriter program could be protected by copyright. The court ruled that there is a demonstrable connection between the AI model developer's input and the AI's output. Thus, the work can be considered human-created.
By contrast, in the Master of Goldsmiths Case from 2020, the court ruled that the AI-generated content did not qualify as a work under the Copyright Law, as it lacked human involvement.
More recent examples include the Springwind Case from 2023 – the first AI-generated image case in China. Here the court recognised that content created by using a generative AI model – based on the Stable Diffusion model – could be protected under China’s copyright law.
“The courts emphasised the human involvement, efforts and decisions in setting up prompts, parameters, selecting from interim outputs and processing further. Thus, the copyright was confirmed,” says Jiang.
“This does not automatically mean that AI-generated copyright can be protected in China. It needs to be evaluated on a case-by-case basis. As AI and our understanding of the technologies develops, we may see different approaches and results,” she observes.
Dealing with infringement risks
Like in other jurisdictions, the risk of infringement with AI relates to both using copyright material to train models, and generating content that it deemed too similar to already copyright-protected material.
Under Chinese law, using copyrighted date to train AI models would constitute “reproduction” and “use” of copyrighted works. The current law does not yet provide explicit rules or fair-use defence to exempt data training from copyright infringement.
“Therefore, theoretically there is still a high legal risk of doing such data training in China. That being said, there is court case ongoing that involves the data-training issue, so will see how the judicial practice looks at this,” observes Jiang.
“We do have voices coming out from industry and scholars discussing if we should give exemptions for using copyrighted works in data training for AI models. We may see developments of the law and we should continue to monitor this,” she says.
When it comes to infringement by AI generated content, the Guangzhou Internet Court found AI platform Tab to be liable for violating the reproduction and adaptation rights of the Ultraman superhero franchise.
The court emphasised that AI service providers must follow regulatory measures, including establishing reporting mechanisms, notifying users of potential risks, and labelling AI-generated content. The defendant failed to implement these measures, resulting in liability for civil damages.
“It’s important to note that the court was examining the liability of the AI service provider in this case – not that of the developer. The court was basically saying that the provider should establish mechanisms to notify users about the potential risk of generating certain content,” explains Jiang.
How can IP owners and AI developers protect themselves?
With AI policy and legislation evolving in China as much as in other markets, Rouse recommends the following principles for staying on top of this exciting new technology:
