When AI and Copyright clash: Chinese courts find AI platform liable for contributory copyright infringement based on AI generated images

Factual background of the case 

The unnamed AI platform involved as the defendant in this case operated an AI platform offering Low-Rank Adaptation (LoRA) models, enabling users to generate images and conduct online model training. The platform in question particularly featured AI-generated images and LoRA models related to "Ultraman", which is a famous Japanese science fiction media franchise. The Ultraman LoRA models were promoted on the homepage of the AI platform, listed specifically under "Recommendations" and "IP Works", and allowed users to apply, download, and publish or share models. This in turn allowed other users to generate pictures that were substantially similar to the image of the well-known Ultraman by entering prompts, selecting basic models and superimposition of the pre-made Ultraman LoRA models. 

The Parties’ arguments

The arguments brought by the parties can be summarized as follows. 

The claimant essentially argued that the AI platform infringed its copyrights (and particularly its information network dissemination rights) in the Ultraman character and engaged in unfair competition by using generative AI to create and distribute infringing Ultraman LoRA models and images, seeking an injunction and RMB 300,000 (approx. 42.000) in damages. After the first instance judgment considered that the AI platform had committed contributory copyright infringement but rejected the unfair competition claims, and granted only RMB 30,000 (approx. USD 4200) in damages and associated costs, the claimant appealed the case to the Hangzhou Intermediate People's Court. 

In response, the AI platform essentially contended that it was merely acting as a network service provider, and its AI platform only integrated third-party uploaded open-source models without providing training data, prompts or original images, relying on the platform’s users to upload materials for training, and thus falls under the "safe harbor" exemption, shielding it from infringement liability. 

The first instance judgment by the Hangzhou Internet Court

The court first analyzed the allegations of copyright infringement. As to direct copyright infringement, the court considered that while an AI platform can in theory be liable for direct copyright infringement if it directly engages in acts that are reserved to the copyright owner (e.g. information network dissemination of copyrighted works), there was no evidence in this case that the platform was the direct provider of allegedly infringing content and was not directly engaging in behavior controlled under the information network dissemination rights, thereby excluding direct copyright infringement here. 

Turning to contributory copyright infringement, the court considered a number of elements in detail: 

• The AI service’s nature provided by the AI platform: The court highlighted that the defendant, as a commercial AI service provider, directly modifies open-source models for user-specific applications, generates AI content for profit, and earns income through membership fees and incentives. It therefore should maintain sufficient understanding of the content in specific application scenarios and bear the corresponding duty of care. 
• Ultraman’s popularity and the obviousness of the infringing nature: the court considered that the Ultraman series is a widely recognized character protected by intellectual property, making infringement easily identifiable. The platform actively featured infringing images among its homepage recommendations and categories, meaning that the presence of unauthorized Ultraman content was obvious and should have been addressed. 
• Reliability and stability of generation of infringing Ultraman images with models, unlike in normal AI-generation scenarios: the court considered as one of the key factors that the platform’s AI technology made it possible to produce highly accurate reproductions of Ultraman by using the LoRA models. Unlike standard AI-generated images, which are generally random and unpredictable, the superimposition of a specific LoRA model allowed for the consistent recreation of Ultraman's original visual characteristics. The court found that the provision of this level of technological precision and stability of output increased the platform's responsibility to oversee user-generated content. In other words, the AI platform should have foreseen the possibility of copyright infringement occurring during the provision of its generative AI services. 
• The AI platform’s profit model: The AI platform generated revenue through users recharging membership and points, and set reward measures to encourage users to publish training models, share links, and invite friends. The court considered that the defendant directly benefited economically from the creative services provided by the AI platform. 
• AI platform’s passive attitude: the court also noted that the AI platform took reactive rather than proactive measures to prevent the obvious copyright infringement. In the platform user service agreement, the AI platform stated that it would not review the content uploaded and published by users, and the evidence did not clearly show that the AI platform had a complaint and reporting channel in a relatively obvious position on its website. Although the platform subsequently blocked some infringing content and introduced background intellectual property reviews, these actions only occurred after the case was already brought before the court. The court determined that the defendant thereby showed that it had the capability to implement preventive measures earlier, but failed to do so. 

Based on the above elements, the court considered that the AI platform could not benefit from the safe harbor exemption. The court concluded that the AI platform should have been aware of its users' copyright infringement, but failed to take necessary preventive measures, demonstrating a lack of reasonable care and thereby aiding in the infringement. The court also concluded that further violations of the Anti-Unfair Competition Law were unnecessary to consider, as it already concluded that the AI platform had violated the Copyright Law. 

The court therefore ordered the AI platform to cease the infringing activities and pay the copyright owner RMB 30,000 (approx. USD 4200) in damages and associated costs. The court order included an order to delete the infringing images and LoRA models involved in the case. Furthermore, the court ruled that the defendant was not required to delete all user-uploaded content, but only materials directly infringing the plaintiff's rights, and denied the plaintiff's request for broader data removal. 

Interestingly, the court also pointed out that images created by AI technology could in theory be covered by the fair use defense in the absence of evidence that the technology uses the original elements of the work and affects its normal use or unreasonably damage the legitimate interests of the copyright owner. 

The second instance judgment 

The court mainly refuted the claimant’s claim first brought in the first instance that the AI platform's behavior constituted unfair competition by discussing the legislative value of the Anti Unfair Competition Law and the concept of technology neutrality.

• The “specious” pictures on the AI platform: the claimant claimed in the first instance procedure that the defendant, through training, generating, and publishing the aforementioned content on an AI platform, had committed unfair competition, citing 18 images and models as being somewhat specious to the image of Ultraman's works, In response, the court held that the claimant's claim was based on the copyright enjoyed by the rights holder based on the Ultraman series of works, and should still be assessed under the Copyright Law. The Anti Unfair Competition Law, as a supplementary protection law, should not repeatedly evaluate infringement if infringement has already been found under the Copyright Law. 
• The nature of the AI services: The IP owner of the Ultraman series also claimed that the generation and publishing activities provided by the AI Platform to users for targeted training on Ultraman constituted unfair competition. However, the court determined that the AI platform only generated works or models related to Ultraman when specific users used specific graphic and textual materials related to Ultraman. Therefore, it did not agree that the AI services were a kind of targeted training. The evidence in the case was also insufficient to prove that the AI Platform had the subjective intent to violate the principles of good faith and recognized business ethics, obtain improper competitive advantages through the AI platform, or harm the legitimate rights and interests of other operators. 
• The responsibilities of the AI platform: In evaluating the AI Platform’s behavior, the court considered what would constitute the appropriate AI platform responsibilities, requiring the platform to timely identify and remove potentially infringing content by setting reasonable terms of use, user guides, and effective monitoring mechanisms, among other necessary measures, to prevent the generation and dissemination of illegal content. In this case, the court held that the AI platform had launched a series of measures after the case was brought before the court to reduce possible liability, demonstrating its proactive attitude towards potential problems, indicating that its business model itself does not intend to disrupt market competition order or gain unfair competitive advantages. 
• Outcome: In summary, considering that the defendant did not subjectively violate the principles of good faith and business ethics, objectively did not engage in any behavior of obtaining improper competitive advantages or harming the legitimate rights and interests of other operators, and took positive preventative control measures, and did not disrupt the market competition order, the appeal court ruled that the AI platform’s behavior did not constitute unfair competition in addition to the indirect copyright infringement which had already been found in first instance. 

The Implications of this case

These judgements are some of the first judgments of their kind in which a court explicitly assesses whether an AI platform can commit copyright infringement by the output it generates and is therefore considered trailblazing. The case underscores the evolving legal landscape concerning AI-generated content and may serve as an indication that the Chinese courts may move towards the general view that AI platforms facilitating generative AI must exercise due diligence and fulfil a duty of care to avoid contributory copyright infringement, even if they do not directly engage in infringing activities such as the direct dissemination or direct creation of the infringing works at issue. 

Specifically, considering the current efficiency of information dissemination, it seems like the Chinese courts tend to raise the bar for the standard of duty of care for AI platforms that "know or should know" about unlawful activity. However, their duty of care should be dynamically and proportionally adjusted according to their information management capabilities. As emphasized by the second instance court, the platform's fault determination should be simply limited to the application of relevant provisions such as the "safe harbour" exemption and the "red flag" principle, but needs to balance the relationship between copyright protection and AI technology innovation and development. In fact, the claimant in this case has also initiated a civil lawsuit in Guangdong Province against another AI platform for generating allegedly infringing content (case number: (2024) Yue 0192 Min Chu 113), in which the court has also imposed specific reasonable care obligations on the AI platform similar to the defendant in this case, including requiring it to establish a complaint reporting mechanism, alerting users to potential copyright infringement risks, prominently labelling/watermarking AI generated content, etc. In short, this case seems to further indicate that the AI platforms should find a balance between technological neutrality and infringement liability, neither excessively restricting technological innovation, nor allowing infringement to occur, with the fulfilment of their duty of care as the proposed red line. 

Additionally, the classification and hierarchical division of AI generated content's infringement liability is also one of the main points of inspiration brought about by this case. In this case, multiple parties are related to the infringing AI generated content, including the AI tool developer, AI service provider (AI platform), providers of the training data, and users of the AI platform. However, the court did not clearly comment on the division of AI generated content's infringement liability between the AI platform and other parties involved. In fact, depending on the architecture of the AI technology, the mode of platform service provision, the characteristics of source data, and the way the users use it, the exact proportion and origin of AI generated content's originality are different, and the specific application scenarios of AI may be ever-changing. For instance, when users are deeply involved in content generation, it should be unreasonable to assume that the copyright in AI generated content is owned by the users, but at the same time to pursue AI platforms for the majority of AI generated content's infringement liability. In addition, this case does not yet involve the issue of commercial large-scale use of AI tools to generate content that infringes on the copyright of others. If it were in the context of commercial large-scale use, the division of infringement liability from AI generated content may result in a different outcome from the outcome reached in this case. 

Finally, the case also reflects the broader trend of Chinese courts taking the international front stage in resolving ground-breaking types of IP disputes. So far, it has been especially the internet courts, established to handle internet-related disputes, playing a pivotal role in shaping the legal framework governing artificial intelligence and other innovations in the online sphere. It will be interesting to follow these developments and we will keep you duly updated. 

Authored by Stefaan Meuwissen, Bowen Chen, and Helen Xia.