This decision establishes a precedent for AI-generated content liability in China, confirming that AI platforms cannot evade responsibility for copyright infringement by blaming users.

Platforms that actively facilitate the distribution of AI-generated infringing content may be held liable for contributory infringement.

Hangzhou Internet Court: Generative AI Output Infringes Copyright

by ChatGPT-4o

The case involves the Hangzhou Internet Court ruling on whether an AI platform is liable for copyright infringement when users generate images based on copyrighted characters.

Case Overview

• The plaintiff (presumably Tsuburaya Productions, the rights holder of Ultraman) sued an AI platform for allowing users to generate images resembling Ultraman using AI-powered LoRA (Low-Rank Adaptation) models.

• The defendant operated an AI platform that provided image generation tools, including the ability to train and apply LoRA models based on user-uploaded data.

• The AI-generated images, trained using Ultraman references, were accessible on the platform and could be downloaded, shared, and further modified by users.

• The plaintiff argued that the platform facilitated copyright infringement and unfair competition by enabling users to train AI models with copyrighted material.

Key Arguments

Plaintiff's Arguments

1. Direct and Contributory Infringement

   • The AI platform hosted LoRA models trained on Ultraman images, allowing users to generate images that closely resembled the copyrighted character.

   • By making these AI-generated images and models available, the platform facilitated the distribution of infringing content, violating information network dissemination rights.

2. Unfair Competition

   • The platform benefited commercially from infringing activities, allowing users to generate and share copyrighted content without authorization.

   • This gave the platform an unfair advantage over legitimate businesses that respect IP rights.

Defendant’s Arguments

1. Safe Harbor Defense

   • The platform did not provide training data but merely offered AI models that users could train on their own data.

   • It only provided AI-based tools and should not be held liable for user-generated content, arguing that it falls within the "safe harbor" provisions for online service providers.

2. No Direct Infringement

   • The platform did not itself create infringing images or explicitly promote their creation.

   • No evidence showed that the platform jointly provided infringing works with users.

Court’s Ruling

1. AI Platform’s Liability for Contributory Infringement

   • The platform had knowledge of infringing activities since Ultraman images were easily identifiable in its recommendations and categories.

   • The platform directly profited from the infringing use, through membership fees and incentives for users to create and share trained models.

   • The court found that the platform should have foreseen the risk of copyright infringement and had an obligation to prevent it.

   • Verdict: The platform was held liable for contributory infringement and ordered to cease the infringing activities and pay damages of 30,000 RMB.

2. Unfair Competition Claim Rejected

   • The court ruled that copyright infringement alone does not automatically constitute unfair competition.

   • The platform’s business model itself was not deceptive or dishonest, and therefore, it did not violate China’s Anti-Unfair Competition Law.

3. Platform’s Duty of Care

   • The platform failed to take reasonable measures to prevent infringement.

   • While AI services can be neutral, the specific implementation of LoRA models increased the risk of repeat infringement.

   • The court emphasized that AI companies must balance innovation with IP compliance, ensuring that their platforms do not encourage large-scale copyright violations.

Implications of the Ruling

• This decision establishes a precedent for AI-generated content liability in China, confirming that AI platforms cannot evade responsibility for copyright infringement by blaming users.

• Platforms that actively facilitate the distribution of AI-generated infringing content may be held liable for contributory infringement.

• AI service providers must implement stricter oversight mechanisms, such as:

  • Filtering copyrighted content during training.

  • Implementing stronger content moderation to prevent the spread of infringing AI-generated works.

  • Providing clear guidelines to prevent users from uploading copyrighted material.

Conclusion

This case represents a significant legal milestone for AI and copyright law, reinforcing the responsibility of AI platforms to ensure their services do not enable large-scale infringement. The ruling suggests that AI-generated content is not exempt from copyright laws, and platforms must take proactive steps to prevent their technology from being used for unauthorized reproduction of copyrighted works.