Generative AI with Chinese Characteristics: Exploring the Chinese Courts' Approach to Generative AI and Copyright Protection

Written by: Edward Berry

Edited by: Mike Yang

Igor Omilaev - Unsplash

Introduction

The People's Republic of China ("China") has been observed as a 'leading power' [1] in Artificial Intelligence ("AI") and credited with introducing 'the first legally binding instrument in the world specifically aimed at regulating generative AI' [2] ("GenAI").  This article will explore the approach taken by Chinese courts to address the tension between the use of GenAI and the protection of copyright holders. Section 1 will begin by defining GenAI and explaining how the rapid rise of GenAI has created a tension with copyright protection. Section 2 will then look at the developing case law, noting that the Chinese courts have addressed two key issues concerning copyrightability and GenAI. This article will conclude by looking at the wider implications of these decisions, noting that the Chinese courts have begun to establish a framework that can be applied to the evolving legal landscape of GenAI. 

Section 1 - GenAI and Copyright: Generating Tension 

GenAI can be defined as an AI that, following specific prompts from a user, can generate a broad range of outputs, including but not limited to image, text, audio and video. Users of GenAI tools (such as ChatGPT and DeepSeek) can prompt the GenAI tool to generate an output that reflects the user's requirements. The legal right of copyright, which protects an individual's exclusive right over a wide range of original work, could be infringed if a user generates an output through GenAI that closely resembles copyrighted material. To generate an output, GenAI tools are trained on vast amounts of data, which 'may include or consist of copyright-protected works and other protected subject-matter' [3]. The generative capabilities of GenAI and the data that GenAI models are trained on thus 'arouses concerns and disquiet as far as protection of copyright' [4]. These concerns reflect a broader objective that is not unique to China but across jurisdictions: the need to balance technological advancement and innovation in AI with the protection of existing intellectual property rights. Further, the development of GenAI has raised fresh copyrightability conundrums, such as whether a user can hold copyrights in an output that has been generated through GenAI. Whilst the tension between GenAI and copyright protection will be examined within the relevant copyright laws of each court's jurisdiction, the following section will explore how the Chinese judicial system has sought to address this tension. 

Section 2 - GenAI, Copyright & Case Law: Generating a Framework?

The tension between the use of GenAI and the protection of copyright holders has been addressed by the Chinese courts in a series of judgments. In particular, the Chinese courts have shed light on two important issues raised by GenAI in relation to copyright infringement.  The first issue is whether a generated image, which has been generated by a user through a GenAI tool, can infringe on the rights of a third-party copyright holder. The second issue is whether a user can receive copyright protection for a work that they have generated through the use of a GenAI tool.  Although the cases discussed below do not set a precedent (which can only be set by the Supreme People's Court) the reasoning in such cases remain useful reference points for building a framework to address copyright challenges brought by GenAI. Furthermore, it moves China towards establishing a framework of rules that can be used to navigate the challenges and nuances raised between GenAI and copyright infringement. 

2.1 - Can images generated by GenAI infringe on the rights of a third-party copyright holder? - The "Guangzhou Ultraman Case"[5] and "Hangzhou Ultraman Case"[6]

Ultraman is a popular superhero series and is a copyright protected work of the plaintiff in both cases. The Guangzhou Ultraman Case has been described as 'the first in the world to render a judgment on whether a generative artificial intelligence service provider should bear liability for the infringement of other people's prior work by its generated content' [7]. The plaintiff argued that the defendant (an AI company), by offering the user an option to input text into a GenAI tool to generate and download images that were substantially similar to Ultraman, infringed on the plaintiff's copyrights including the right of reproduction and the right of dissemination via the internet. The Guangzhou Internet Court agreed with the plaintiff on the existence of infringement and ruled that the defendant, as a provider of a GenAI service, owed a duty of care. The judgment stated that this duty of care arises from pieces of GenAI regulation [8] and consists of three components: the duty to establish a reporting and complaint mechanism; the duty to warn users of the risks of copyright infringement and; the duty to provide clear and prominent labelling in AI-generated content. After examining the defendant's actions (and inaction) against this criterion, it was found that the defendant failed to fulfil its duty of care as a GenAI service provider and in turn was liable for copyright infringement. This judgment can be seen as providing 'clear compliance guidelines for GenAI service providers regarding their duty of care' [9] and provides a framework for GenAI providers in China to operate within.

The Hangzhou Ultraman Case (heard seven months after the Guangzhou Ultraman Case) similarly found that a GenAI service provider could be liable for copyright infringement. The Hangzhou Internet Court made a judgment of first instance applying a duty of care approach to determine copyright liability. The defendant operated a GenAI service that offered Low-Rank Adaption models ("LoRA"), which are GenAI models that allow a user to upload images and train the model on that image. This allowed users to upload images of Ultraman and through input prompts, generate images that closely resembled Ultraman. The plaintiff argued that this infringed on their information network dissemination rights protected under copyright. Unlike the Guangzhou Ultraman Case, the defendant in the Hangzhou Ultraman Case did not directly provide the infringing content and argued that the training data, which contained copyrighted material, was provided by the users. The Hangzhou Internet Court accepted this, noting that the GenAI service provider does not necessarily have to conduct prior review into the training data supplied by the user, finding that the court should not put excessive obligations on GenAI service providers. However, the court found that the defendant failed to take reasonable measures and thus failed to fulfil its reasonable duty of care as a GenAI service provider. To reach this conclusion, the court noted the popularity of the Ultraman character and the identifiable output that the Ultraman LoRA offered users. It further noted that the defendant operated the GenAI service for profit and therefore should have been alert to the possibility of infringement by its users on the facts. Given the failure to take reasonable measures to prevent infringement and the failure to fulfil its reasonable duty of care, the defendant was found liable to aiding and abetting copyright infringement.

It can be argued that these two cases 'provide clear standards for determining the responsibility of AI-related platforms' [10]. Looking at both of these cases, the Chinese courts have taken a duty of care approach that is to be applied on a case-by-case basis. If a GenAI service provider directly provides infringing content to a GenAI model, it is at risk of being found liable for copyright infringement if an output generated by a user closely resembles copyrighted work. To satisfy its duty of care in this instance, it must satisfy the three requirements set out in the Guangzhou Ultraman Case. Alternatively, if a GenAI service provider does not directly provide infringing content to a GenAI model and this content is instead uploaded by the user, the duty of care is satisfied if it takes reasonable measures to minimise copyright infringement by its users. Whilst providing a clear framework for GenAI service providers, this approach affords the Chinese courts with a great amount of discretion to determine whether the duty of care has been met on the facts. This flexibility can give the Chinese courts space to not only adequately address the technological nuances raised in GenAI and copyright disputes, but also to address this tension in light of the rapidly developing GenAI industry. 

2.2 - Can an AI-generated image created through GenAI receive copyright protection? The cases of Li v Liu [11] and Lin XX [12]  (Half Heart Case)

The Chinese courts have handed down two judgments that grant the user copyright over an image generated through GenAI.  In Li v Liu the plaintiff generated an image through the GenAI model Stable Diffusion and added their own watermark to the image. The plaintiff argued that the defendant, by removing the plaintiff's watermark and attaching the AI-generated image to the defendant's article, violated their copyrights, including right of authorship and right of dissemination. In the first case of its kind, the Beijing Internet Court found that an AI-generated image could satisfy the requirements to receive copyright protection under Chinese Copyright Law. To satisfy these requirements, the plaintiff's work must be: (1) an original intellectual creation in the realm of literature, art and science; and (2) expressed in a specific form [13]. The court emphasised the plaintiff's active role in the creation of the generated image, noting the various inputs that the Plaintiff used in the GenAI model to generate the image and thus could be considered an original intellectual creation. The court further analysed the wording of the licencing agreement between the developer of the GenAI model and the user, finding that it stated that the ownership of the output images rested with the user. In light of this, the judgment concluded that the plaintiff held copyrights in the image generated and that the defendant's actions constituted infringement.

In March 2025, the Changshu People's Court similarly found in Lin XX that an AI-generated image can receive copyright protection. The plaintiff used the GenAI tool Midjourney to generate an image of a half-heart, which was then posted on the defendant's website without the plaintiff’s consent and used as inspiration by the defendant to create a three-dimensional object.  The court again looked at the user agreement and found that this agreement stated that the user held ownership rights in images generated by the GenAI tool. The court took a similar approach to Li v Liu in Lin XX, placing emphasis on the plaintiff's role in choosing various inputs and prompts that reflected the user's artistic judgment and decision-making. Through such inputs, the plaintiff was found to have contributed significantly to the creation of the generated image and thus such image was eligible for copyright protection. In both cases the level of human input in the creation of generated images remains a key measurement of whether such image constitutes an 'original intellectual creation' and thus can be eligible for copyright protection in China. The court ruled in Lin XX that whilst the three-dimensional object could not receive copyright protection, the plaintiff's copyrights in the generated image were infringed by the defendant posting such image on their website.   

These two cases demonstrate a willingness of the Chinese courts to grant copyright protection to images generated by a GenAI tool. To receive copyright protection, the court will assess whether the AI-generated image constitutes an 'original intellectual creation' and look at the level of human input in the creation of the image. In both cases, the courts noted that the necessary level of human input will be decided on a case-by-case basis. Nevertheless, a clear focus on the active role of the human and the fruition of their artistic choices provides a clear framework to assess the copyrightability of AI-generated work in China. This builds on other developments in the case law concerning the tension between GenAI and copyright protection. Further, as the first jurisdiction to grant copyright protection to AI-generated works, the reasoning in these cases can serve as useful reference points for other jurisdictions grappling with this issue.

Conclusion- Looking to the Bigger Picture

This article has looked at how the Chinese courts have approached the tension between facilitating the use of GenAI and the need to protect copyright holders. Although the GenAI Industry is at its early-stage of development, much judicial ink has been spilt in China on this tension, with the article discussing two key issues that have been addressed by the Chinese courts. Looking at the case law discussed, a framework of rules has begun to form around the liability of GenAI service providers for copyright infringement of generated outputs and the copyrightability of AI-generated works by users. These judicial practices remain fresh and could be applied, developed and adjusted in future cases, in which new technical arguments are likely to be raised. As this framework develops, the Chinese courts will need to be mindful of the need to protect copyright ownership whilst encouraging innovation and technological advancements in the AI Industry [14]. Whilst many issues raised by the use of GenAI and infringement of copyright are yet to be addressed, it is submitted that the Chinese courts has tackled key questions head-on, resulting in a useful framework of rules that provide guidance for GenAI service providers and users of such services when it comes to navigating this tension. 

REFERENCES

[1],[2] Zou M &, Zhang L,  'Navigating China’s regulatory approach to generative artificial intelligence and large language models' (2025) Cambridge Forum on AI: Law and Governance' 1(8) <https://doi.org/10.1017/cfl.2024.4> 

[3] Rosati E, 'Infringing AI: Liability for AI-Generated Outputs under International, EU, and UK Copyright Law' (2024) European Journal of Risk Regulation 1-25. <https://doi:10.1017/err.2024.72> 

[4] Ballell TLRH, 'Mapping Generative AI rules and liability scenarios in the AI Act, and in the proposed EU liability rules for AI liability' (2025) Cambridge Forum on AI: Law and Governance 1(5) <https://doi.org/10.1017/cfl.2024.8>

[5] Shanghai Xinchuanghua Cultural Development Co Ltd v. an AI Company (2024) Yue 0192 Min Chu No. 113

[6] Shanghai Character License Administrative Co., Ltd v. Hangzhou Jellyfish Intelligent Technology Co., Ltd (2024)

[7],[9] Jiang F & Wu H, 'The world’s first case involving a generative artificial intelligence: Shanghai Xinchuanghua Cultural Development Co Ltd v AI Company (pseudonym)' 2024 14(4) Queen Mary Journal of Intellectual Property

[8] Specially, the court referred to Article 15, Article 4 and Article 12 of the Interim Measures for the Management of Generative AI Services (2023) ("The Measures")

[10] Ren J, 'Two AI Platform Infringement Cases Triggered by Ultraman' (Lexology, 11 April 2025) <https://www.lexology.com/library/detail.aspx?g=9a34a38b-d979-47f8-9125-f7bc70e68523>

[11] Li v. Liu, (2023 Jing 0491 Min Chu No. 11279.)

[12] Lin Chen v. Hangzhou Gaosi Membrane Technology (2024)

[13] See Article 3 of Copyright Law of the People's Republic of China

[14] He X & Shan P, 'China’s regulations on the attribution of AI-generated content: an exploration based on the open-ended approach' (2024) Journal of Intellectual Property Law & Practice, DOI: <https://doi.org/10.1093/jiplp/jpae109>