China: First AI output copyright infringement case

The Guangzhou Internet Court found a GenAI service provider liable for copyright infringement on 8 February 2024. 

While contentious AI issues are growing globally, this was China’s first judgment concerning copyright infringement by AI-generated content.

This landmark ruling follows the issuance of China’s first’s GenAI regulation last August (GenAI Measures), and a separate decision by the Beijing Internet Court in November 2023 where the court recognised an AI-generated image as a copyright work in a case involving its unauthorised reproduction.

Violation of right of reproduction and right of adaptation 

The claimant, a Shanghai-based character licensing agency for well-known Japanese animation shows, is the exclusive licensee of the Ultraman series in China. It claimed that its rights to the Ultraman works were infringed by the defendant’s text-to-image AI service, with identical and substantially similar images to Ultraman works being generated with prompts containing or related to “Ultraman”.  

The court ruled in favour of the claimant, citing the following infringements:

• Right of reproduction: the images generated by the service are indeed either identical or substantially similar to the original Ultraman works.  
• Right of adaptation: certain images generated by the service partially maintained the original expressions of the Ultraman works, while incorporating new features which constituted unauthorised derivative works of the Ultraman originals. 

Application of the GenAI Measures

The court also took the view that the defendant, which falls under the definition of a GenAI service provider under the GenAI Measures, had not complied with these new regulations. Specifically, the defendant failed to adhere to the following requirements imposed on GenAI service providers: 

• Article 15: An online complaint and reporting mechanism must be set up for rights holders, with information on the procedures and timeline involved. 
• Article 4: Risk warnings must be included in terms of use or service agreements to remind users to respect intellectual property rights of others.
• Article 12: AI-generated outputs must be prominently labelled to prevent confusion. 

A step forward for GenAI content 

This case highlights China’s commitment to protecting intellectual property rights in the digital age. GenAI service providers should keep abreast of the rapidly evolving intellectual property and GenAI regulatory landscape. It will be imperative that they take proactive measures to fulfil their duties of care while mitigating potential IP infringement risks. The adoption of the complaint mechanisms, risk warnings, and AI labelling measures are obvious starting points given the court’s findings above.

In addition, although training data compliance was not a specific focus tackled by the court (as the GenAI model was not trained by the defendant itself in this case), GenAI service providers should also be aware of potential IP infringement risks when training their AI models.

A recent technical guidance released by China’s standard committee outlines a helpful checklist below:

• Appoint a responsible person for IP-related matters concerning training data and generated content, and to establish an IP management strategy.
• Identify IP risks before using the data sources for data training purposes. Special attention should be given to copyright infringement when the training data includes literary, artistic, or scientific works.
• Set up a channel for complaints and reports on IP issues.
• Inform users, via a user agreement, of the IP risks associated with using the generated content and agree with the users on the responsibilities and obligations for identifying IP issues.
• Keep track of the national policies and third-party complaints to keep IP strategies up-to-date.
• Where appropriate, disclose IP information about the relevant training data. Support third parties in the complaint and report channels to inquire about the use of the training data and related IP issues.