Liability of AI Service Providers for Copyright Infringement: Guangzhou Internet Court reaches world’s first decision

However, following a decision last month, the Chinese courts have beaten the rest of the world to reaching a decision regarding the liability of AIGC for copyright infringement. Although the case does not address the use of copyrighted training data and focus rather on the liability of the outputs, China is the first to address the legal issues of generative AI and is starting to become a reoccurring theme. While the EU’s AI Act was once the frontrunner in the global race to regulate generative AI, it has been beaten into second place by China’s interim measures on generative AI, which came into force in August 2023 (listen to our podcast comparing the two approaches here [https://www.twobirds.com/en/insights/2023/global/regulating-ai-%E2%80%93-perspectives-from-europe-and-china]).

In February 2024, the Guangzhou Internet Court rendered China's first ruling on AIGC’s liabilities in a case involving AI-generated images of Ultraman as prompted by the plaintiff. The first instance ruling was issued in lighting speed: just one month and three days after case filing.  The three internet courts located in Beijing, Hangzhou and Guangzhou have jurisdictions over copyright infringement which occurs online and usually takes only a few months to reach a judgment in comparison with the time of civil litigation which is around 6 months to -2 years.

The plaintiff in this case was Shanghai Character License Administrative Co., Ltd. (SCLA), which has exclusive rights to the Ultraman series (including various rights such as reproduction, adaptation, and information network communication) granted by Tsuburaya Productions and the authorization to claim the licensed rights in litigations. The defendant was an undisclosed AI Company that operates a website (anonymised as “Tab”) with AI conversation function and AI image generation function. These functions were provided using an unnamed third-party provider’s AI model which was connected via a programmable interface (API) to the defendant’s website. While the defendant was allowing visitors to their website to use this AI model to generate images, they had not created the AI model themselves. SCLA discovered that when the defendant’s website was asked to generate Ultraman-related images (e.g. through the prompt "generate an Ultraman Dyna"), the image generated by the website bore a resemblance to Ultraman. SCLA claimed copyright infringement ]and requested the Court to order the defendant to cease infringement and remove the Ultraman artistic images from the defendant’s training dataset. It also requested damages awards and reasonable costs of RMB 300,000. The Court awarded RMB 10,000. The highlights of the judgment are as follows:

Court Decision:

The Guangzhou Internet Court issued a first-instance judgment on whether a provider of generative AI services infringed upon the plaintiff's reproduction and adaptation rights when offering generative AI services. The judgment concluded that the Ultraman figures, as original artistic creations, are protected under the Copyright Law of the People’s Republic of China and the images generated by the defendant's website partially or fully resemble Ultraman figures. The Court concluded that this constituted an infringement of the plaintiff's rights of reproduction and adaptation  and the defendant was held liable for infringement. The case also cites various administrative regulations on the duty of care that generative AI service providers should exercise. Whether or not the duty of care had been complied with impacts on the damages assessment.

Rights Infringed:

The plaintiff SCLA claimed copyright infringement of reproduction right, adaptation right, and information network communication right. The court carefully compared the images generated by the defendant’s website with the copyrighted Ultraman artistic figures to ascertain whether the claimed copyrights were infringed.

a. Infringement of Reproduction Right:

The court determined that the defendant did infringe upon the reproduction right of the Ultraman work, as the AI-generated images substantially replicated the original expression of the “Ultraman” artistic figures without permission, and the defendant had likely accessed the well-known Ultraman artistic figures before.

b. Infringement of Adaptation Right:

Under Chinese law the adaptation right prevents others from altering the work without the authorisation of the copyright owner to create a new work of original creation. The court found that the defendant had infringed upon the adaptation right of the Ultraman artistic figures, as the generated images partially retained the original expression of the work while introducing new features.

c. Infringement of Information Network Communication Right:

The information network communication right prevents others from making the work available to the public by wired or wireless means without the authorisation of the copyright owner, so that the public can access the work at a time and place of their choosing. The court decided not to evaluate this particular right, considering that the infringement of the reproduction and adaptation rights had been established, and also considering that this was a novel case regarding the infringement of a product created in the context of the development of generative artificial intelligence. The court concluded that the alleged infringement has already fell within the scope of control of the reproduction and adaptation rights.

Ceasing Infringement Action:

a. Preventing Generation of Similar Images:

The court dismissed the defendant's argument that its AI painting service was outsourced to a third-party provider whose product is connected via a programmable interface to the defendant’s website. The Court ordered the defendant to implement technical measures such as key words filters to prevent the generation of images substantially similar to the Ultraman artistic figures.

b. Removal of Ultraman Images from Training Dataset:

As the defendant did not actually train the model, the court did not support the plaintiff's request to remove Ultraman materials from its training dataset. However, as the third party was not named as a defendant in this case, the court did not clarify the third party's liability.

AI Provider's Duty of Care:

The court addressed the defendant’s liability for damages, emphasizing the need for service providers to exercise reasonable duty of care. These duties are set out in Article4, Article 12 and Article 15 of the Interim Measures for the Management of Generative Artificial Intelligence Services and include the establishment of a IPR complaint and reporting mechanism, adequate warnings and terms & conditions against the use of its service to infringe third party rights, and conspicuous labelling of AI-generated output. In this case the defendant was found to be at fault and therefore is responsible for compensating the plaintiff.

Compensation Amount:

Considering the high market recognition of the Ultraman series, the defendant's efforts to mitigate further infringements, the limited impact range, and the plaintiff's reasonable expenses for evidence collection, the court ordered a compensation of RMB 10,000 (around $1,400 USD) (including reasonable expenses) to be paid by the defendant to the plaintiff.

Legal and Technological Implications:

This case underscores the challenges and complexities of copyright law in the era of artificial intelligence. The Court’s findings that (i) a similarity between the output of a generative AI system and pre-existing works can amount to an infringement of the rights in those pre-existing works; and (ii) the operator of a website offering that AI system can be liable for those outputs, are both significant findings for generative AI providers and rights holders. The finding that operators can be liable for outputs created using these systems will be of concern, although the low level of compensation awarded and Court’s approach to requiring the implementation of AI governance measures in limiting damages exposure will offer some reassurance that the Chinese courts take a pragmatic approach to resolving these issues.

It is not clear from the decision whether the Court was asked to examine the text prompts which were used to create the relevant outputs. This issue has received a lot of attention in US court proceedings, where some defendants alleged that the plaintiffs are essentially trying to “hack” their generative AI systems by using carefully crafted prompts to make those systems produce outputs which are similar to the plaintiff’s own works.

This case focuses on liability relating to the outputs from a generative AI system and does not address whether using copyright materials in AI training constitutes infringement. It does however provide a useful discussion of the infringement liability and due diligence obligations of AI-generated service providers in China. It is noteworthy that the judgment also highlights the strategic importance of AI as a driving force of technological innovation for China and the need for a balanced legal framework to protect both the copyright and the development of the AI industry. The judgment considers that “it is inappropriate to put excessive obligations on the service providers”. It remains to be seen whether a subsequent case will be launched against the third-party provider, in which the court will be confronted with the equally difficult question on whether the use of copyrighted materials in training data constitute an infringement in China.