Ultraman AI Case in China: Defining Copyright Liability for Generative AI Providers

Written by Dr. Leslie (Liang) Xu IP Expert and collaborator of the China IP SME Helpdesk.

Recently, the Intermediate People’s Court of Hangzhou, Zhejiang Province, delivered a second-instance judgement on a high-profile copyright infringement case involving a generative artificial intelligence platform. The court found that the defendant’s provision of related services constituted contributory infringement of the right of communication through information network, thus dismissing the appeal and upholding the original judgment.

This case, which follows the landmark ruling by the Guangzhou Internet Court in February 2024 — the world’s first infringement case involving a generative AI platform — represents another lawsuit filed by the same plaintiff concerning the same series of works. Compared to the initial case, this judgment provides a more detailed and comprehensive discussion of the legal responsibilities of generative AI service providers, carrying significant judicial reference value.

The plaintiff in this case holds the intellectual property rights to the Ultraman series of images. The defendant operates an AI platform that allows users to upload Ultraman images to train a LoRA model, which then generates new images substantially similar to the plaintiff’s copyrighted works by overlaying a base model. Consequently, the plaintiff claimed that the defendant’s actions constituted an infringement of the right of communication through information network and unfair competition, demanding compensation of CNY 300,000 (approx. 43,000). The first-instance court found that the platform constituted contributory infringement and awarded CNY 30,000 (approx. 4,300) in damages, but dismissed the unfair competition claim. The primary contention in this case revolves around whether the accused conduct infringes upon the right of communication through information network and the determination of civil liability. The court’s reasoning highlights the following noteworthy points:

First, the judge pointed out that, given the technical characteristics of generative AI, liability for infringement should be distinguished and classified according to different application scenarios. The court segmented the process of generative artificial intelligence into different stages, adopting a lenient standard for determining infringement at the data input and training phases, while applying a stringent standard for determining infringement at the content generation and usage phases. On one hand, if the generative AI platform directly performs actions controlled by exclusive copyright rights, it may constitute direct infringement; on the other hand, when users input infringing images and decide whether to generate and publish them, the defendant does not necessarily have an obligation to preemptively review the input training images and generated content. Only when the defendant is at fault for specific infringing actions can it constitute contributory infringement.

Second, in analyzing the platform’s duty of care, the court considered the particularities of disputes involving artificial intelligence. It comprehensively assessed the nature and profit model of generative AI services, the reputation of the copyrighted work, the obviousness of the alleged infringement, the possible consequences of infringement caused by generative AI, and whether reasonable measures were actively taken to prevent infringement.

As a result, the first-instance court ruled that, although the defendant did not directly perform actions controlled by the right to disseminate information on the network, it should have been aware that network users were using its services to infringe on this right yet failed to take necessary measures, clearly neglecting reasonable duty of care, thus constituting contributory infringement of the right of communication through information network. The second-instance court upheld this view.

The significance of this case lies in:

For one, the case deeply explores important legal issues surrounding the particularities of generative artificial intelligence (AIGC) services and their infringement disputes. It clarified the duty of care and fault determination standards for platforms in data training, content generation, and dissemination, delineating the boundaries of liability for AIGC platforms in copyright infringement cases. This provides an important judicial reference and guidance for adjudicating similar cases in China in the future and offers a legal basis for the regulated development of the generative AI industry.

For another, the case highlights the key compliance requirements that AI enterprises need to focus on when operating generative platforms.

1. Duty of care is proportional to profitability: The more profitable a platform is, the higher its duty of care.

2. Technical architecture and copyright recognition: If the platform employs dedicated model training, it must be equipped with more robust copyright recognition systems to mitigate infringement risks.

3. Prevention over remediation: The platform’s response to infringement must be timely and effective; it cannot rely solely on post-incident remedies, as preemptive prevention is paramount.

For further reference, review the following judicial documents:

• First Instance Judgment:（2024）浙0192民初1587号
• Second Instance Judgment:（2024）浙01民终10332号