On February 10, 2025, the Hangzhou Court of the Internet issued a first instance judgment regarding an assisting infringement of the information network broadcasting right by an AI company in Hangzhou, and the said AI company was ordered to cease the infringement immediately and compensate for economic loss, as well as pay a reasonable fee of RMB 30, 000. Currently, the said judgment has come into effect.
[Case introduction]
The Plaintiff of the case is the intellectual property rights owner of the Ultraman series, whereas the Defendant is a company that operates an AI platform that provides Checkpoint basic models and LoRA models and supports many functions such as graph generation and online model training.
It is reported that there were AI-generated Ultraman images and LoRA models on the front page, “Recommendation,” and “IP Work” page of the said platform, which can be applied, downloaded, published, or shared. The Ultraman LoRA model is generated by the user uploading the Ultraman picture, selecting the basic model of the platform, and adjusting the parameters for training. Then, other users may enter prompts, select the basic model, and superimpose the Ultraman LoRA model for training to generate pictures that are substantially similar to the image of Ultraman.
The Plaintiff claimed that the Defendant infringed its information network broadcasting right by placing the infringing image and model in the information network upon training and generating the input images, and such targeted training of Ultraman LoRA models and generating infringed images by the AI technology also constituted unfair competition. Thus, the Plaintiff requested the Defendant cease the infringement and compensate for economic loss RMB 300, 000.
The Defendant replied that the AI platform integrates into a generative AI platform that can be directly applied by users by using third-party open-source model code and carrying out engineering operations such as technology integration and application deployment in combination with the requirements of the platform's use scenarios, but it does not provide training data. Instead, the user feeds the image materials to the model for learning and training to generate images; thus, it should fall within the scope of the platform's liability exemption under the "safe harbor" rule and does not constitute infringement.
[Court opinions]
According to the Hangzhou Court of the Internet, since there was no evidence proving that the defendant and users co-provided the infringing works, the defendant did not directly conduct behaviors that were controlled by the information network broadcasting right. However, it may constitute assistance to infringement when the defendant was at fault regarding specific infringing acts based on the factors as follows:
1) The nature and profit model of the generative AI service: the Court considered that the open-source ecosystem as an important part of the artificial intelligence industry provides a general basic algorithm logic, whereas the Defendant has made targeted modifications and improvements on the basis of the open-source model in combination with specific application scenarios, provided solutions and results that directly meet the needs of users, and directly participated in business practices and benefited based on the content generated by the target. Thus, from the perspective of service type, commercial logic, and prevention cost, it should have maintained a sufficient understanding of the content in specific application scenarios and assumed the corresponding duty of care. In addition, the Defendant obtained income from users recharging memberships and points and set up incentive measures to encourage users to publish training models, etc.; therefore, it was considered that the Defendant directly obtained economic benefits from the creation services provided by the platform.
2) The fame of the copyrighted work and the degree of obvious regarding the claimed infringing facts: the Court recognized that the Ultraman works had retained comparatively high fame, and there had been multiple infringing images shown on the front page and specific classification displays on the platform. Further, the Court considered that the direct display of the infringing images by the LoRA model cover or example images fell into the category of the infringing information that can be obviously sensed.
3) The possible results by the generative AI: The Court considered that due to the superimposition of the Ultraman LoRA models, the generative AI herein can stably output the characteristics of the Ultraman images, which enhanced the platform’s recognizability and interventionability of the results of user behavior. Further, since the images generated by users and the LoRA models can be repetitively used by other users due to the convenience of the technology, the situation of causing the spread of infringement is quite obvious. Thus, the Defendant should have foreseen the possibility of infringement occurring.
4) Whether the reasonable measures to prevent infringement are positively adopted: the Defendant claimed in its user service agreement that it shall not review the content uploaded and published by the users but blocked the relevant content upon the receipt of the litigation notice and adopted measures to conduct IP right review at the backstage. The Court considered that such acts proved that it had the ability but failed to adopt necessary measures that are in line with the technology level when the infringement happens to prevent the infringement.
In a word, the Court considered that the Defendant should have known the network users took advantage of its service to infringe the information network broadcasting right but failed to take necessary measures, as well as the duty of reasonable care, and it was subjectively at fault. Thus, the Defendant has constituted assistance to infringement.
In terms of unfair competition, the Court reasoned that the platform service has not violated the principle of honesty and credit, as well as commercial morality, since it aims to enlarge the application and function of the generative AI and provide more individual creation service to the users. If the users can create content while respect others’ IP rights under the platform service agreement, it should not have infringed the rights of the copyright owner, nor the social public interest. In addition, the Court considered that in case the generation of the AI reaches the level of reproducing the original expression of others’ work, it falls within the scope of copyright laws; the Anti-unfair Competition Law as the supplement to the Copyright Law shall not repeatedly comment on such infringement. In a word, the Defendant has not constituted unfair competition.
Regarding the civil liability, the Court considered that the defendant shall immediately delete the generated and published infringing images, as well as the infringing LoRA models that can display the original design features of the copyrighted work, and take measures to effectively prevent the infringement. The Court also stated that if there is no evidence to prove that the generative AI is for the purpose of using the original expression of the copyrighted work or has affected the normal use of the copyrighted work, or unreasonably harmed the legitimate interests of the relevant copyright owner, it can be considered as fair use. Thus, when the user can still study, research, and enjoy the relevant pictures stored in the platform, or make other reasonable use of the pictures and has not disseminated them to the outside world, or the right holder or its authorized person uses the relevant pictures on their own, the plaintiff's general request on the defendant’s deletion of all materials and relevant data related to its copyrighted work was not supported by the Court.
[Comment]
In this case, the Court did not recognize the AI platform as a direct infringement but as an assistance to infringement as it failed its duty of reasonable care. This is helpful to draw the lines of the platform’s responsibility; that is, the platform is liable for taking measures (e.g. reviewing the content generated by users, etc.) to prevent the infringement even though it does not directly participate in such infringement.
It is also noticed that the Court tried to balance the interest between technology innovation and the intellectual property protection. For instance, the Court recognized that “the technology itself is neutral,” and the commercial and operational model of the AI platform is to “expand the application scenarios and functions of generative AI to provide users with more personalized creation services, improve creative efficiency, and not violate the principles of integrity and business ethics.” Thus, it shall not deny the application of the AI technology but draw the line when the platform should put on more responsibility to avoid infringing on others’ rights. No word as to whether any appeal has been filed, have been received so far.
[Case introduction]
The Plaintiff of the case is the intellectual property rights owner of the Ultraman series, whereas the Defendant is a company that operates an AI platform that provides Checkpoint basic models and LoRA models and supports many functions such as graph generation and online model training.
It is reported that there were AI-generated Ultraman images and LoRA models on the front page, “Recommendation,” and “IP Work” page of the said platform, which can be applied, downloaded, published, or shared. The Ultraman LoRA model is generated by the user uploading the Ultraman picture, selecting the basic model of the platform, and adjusting the parameters for training. Then, other users may enter prompts, select the basic model, and superimpose the Ultraman LoRA model for training to generate pictures that are substantially similar to the image of Ultraman.
The Plaintiff claimed that the Defendant infringed its information network broadcasting right by placing the infringing image and model in the information network upon training and generating the input images, and such targeted training of Ultraman LoRA models and generating infringed images by the AI technology also constituted unfair competition. Thus, the Plaintiff requested the Defendant cease the infringement and compensate for economic loss RMB 300, 000.
The Defendant replied that the AI platform integrates into a generative AI platform that can be directly applied by users by using third-party open-source model code and carrying out engineering operations such as technology integration and application deployment in combination with the requirements of the platform's use scenarios, but it does not provide training data. Instead, the user feeds the image materials to the model for learning and training to generate images; thus, it should fall within the scope of the platform's liability exemption under the "safe harbor" rule and does not constitute infringement.
[Court opinions]
According to the Hangzhou Court of the Internet, since there was no evidence proving that the defendant and users co-provided the infringing works, the defendant did not directly conduct behaviors that were controlled by the information network broadcasting right. However, it may constitute assistance to infringement when the defendant was at fault regarding specific infringing acts based on the factors as follows:
1) The nature and profit model of the generative AI service: the Court considered that the open-source ecosystem as an important part of the artificial intelligence industry provides a general basic algorithm logic, whereas the Defendant has made targeted modifications and improvements on the basis of the open-source model in combination with specific application scenarios, provided solutions and results that directly meet the needs of users, and directly participated in business practices and benefited based on the content generated by the target. Thus, from the perspective of service type, commercial logic, and prevention cost, it should have maintained a sufficient understanding of the content in specific application scenarios and assumed the corresponding duty of care. In addition, the Defendant obtained income from users recharging memberships and points and set up incentive measures to encourage users to publish training models, etc.; therefore, it was considered that the Defendant directly obtained economic benefits from the creation services provided by the platform.
2) The fame of the copyrighted work and the degree of obvious regarding the claimed infringing facts: the Court recognized that the Ultraman works had retained comparatively high fame, and there had been multiple infringing images shown on the front page and specific classification displays on the platform. Further, the Court considered that the direct display of the infringing images by the LoRA model cover or example images fell into the category of the infringing information that can be obviously sensed.
3) The possible results by the generative AI: The Court considered that due to the superimposition of the Ultraman LoRA models, the generative AI herein can stably output the characteristics of the Ultraman images, which enhanced the platform’s recognizability and interventionability of the results of user behavior. Further, since the images generated by users and the LoRA models can be repetitively used by other users due to the convenience of the technology, the situation of causing the spread of infringement is quite obvious. Thus, the Defendant should have foreseen the possibility of infringement occurring.
4) Whether the reasonable measures to prevent infringement are positively adopted: the Defendant claimed in its user service agreement that it shall not review the content uploaded and published by the users but blocked the relevant content upon the receipt of the litigation notice and adopted measures to conduct IP right review at the backstage. The Court considered that such acts proved that it had the ability but failed to adopt necessary measures that are in line with the technology level when the infringement happens to prevent the infringement.
In a word, the Court considered that the Defendant should have known the network users took advantage of its service to infringe the information network broadcasting right but failed to take necessary measures, as well as the duty of reasonable care, and it was subjectively at fault. Thus, the Defendant has constituted assistance to infringement.
In terms of unfair competition, the Court reasoned that the platform service has not violated the principle of honesty and credit, as well as commercial morality, since it aims to enlarge the application and function of the generative AI and provide more individual creation service to the users. If the users can create content while respect others’ IP rights under the platform service agreement, it should not have infringed the rights of the copyright owner, nor the social public interest. In addition, the Court considered that in case the generation of the AI reaches the level of reproducing the original expression of others’ work, it falls within the scope of copyright laws; the Anti-unfair Competition Law as the supplement to the Copyright Law shall not repeatedly comment on such infringement. In a word, the Defendant has not constituted unfair competition.
Regarding the civil liability, the Court considered that the defendant shall immediately delete the generated and published infringing images, as well as the infringing LoRA models that can display the original design features of the copyrighted work, and take measures to effectively prevent the infringement. The Court also stated that if there is no evidence to prove that the generative AI is for the purpose of using the original expression of the copyrighted work or has affected the normal use of the copyrighted work, or unreasonably harmed the legitimate interests of the relevant copyright owner, it can be considered as fair use. Thus, when the user can still study, research, and enjoy the relevant pictures stored in the platform, or make other reasonable use of the pictures and has not disseminated them to the outside world, or the right holder or its authorized person uses the relevant pictures on their own, the plaintiff's general request on the defendant’s deletion of all materials and relevant data related to its copyrighted work was not supported by the Court.
[Comment]
In this case, the Court did not recognize the AI platform as a direct infringement but as an assistance to infringement as it failed its duty of reasonable care. This is helpful to draw the lines of the platform’s responsibility; that is, the platform is liable for taking measures (e.g. reviewing the content generated by users, etc.) to prevent the infringement even though it does not directly participate in such infringement.
It is also noticed that the Court tried to balance the interest between technology innovation and the intellectual property protection. For instance, the Court recognized that “the technology itself is neutral,” and the commercial and operational model of the AI platform is to “expand the application scenarios and functions of generative AI to provide users with more personalized creation services, improve creative efficiency, and not violate the principles of integrity and business ethics.” Thus, it shall not deny the application of the AI technology but draw the line when the platform should put on more responsibility to avoid infringing on others’ rights. No word as to whether any appeal has been filed, have been received so far.