China - Court Decides Artificial Intelligence Generated Content Infringes Copyright
Published 26 March 2024
Xia Yu
On 8 February 2024, China’s Guangzhou Internet Court (“Court”) issued a judgment (2024) Yue 0192 Civil Judgment of First Instance No. 113, which is considered to be one of the world’s first court issued judgments on copyright infringement of artificial intelligence-generated content (“AIGC”). According to the judgment, the Court held a Chinese AI company had infringed on another’s copyrights in the process of providing generative artificial intelligence (“AI”) services and should bear relevant civil liabilities.
The Ultraman image is a world-renowned IP work and has wide influence and popularity internationally. The copyright holder of Ultraman signed an authorization letter with Shanghai Xinchuanghua Cultural Development Co., Ltd. (“Plaintiff”), exclusively authorizing it the copyright of the Ultraman series of images in China (excluding Hong Kong, Macau, and Taiwan). The Plaintiff found that an AI company (“Defendant”) provided an AI painting service on its website. Member users of the Defendant can use the AI painting service by consuming the computing power purchased by their recharge. Through this service, when a member user requests to generate a picture related to Ultraman, the image in the picture generated by the website is substantially similar to the Ultraman image of The Plaintiff. The works generated by the AI painting service provided by the Defendant are mainly based on large amounts of data training, and the training data are a large number of original works, including the Ultraman work of the Plaintiff. The Plaintiff believed that the Defendant had committed online infringement. Thus, in early January 2024, it filed a lawsuit with the Court, requesting the Defendant to stop generating infringing Ultraman pictures, delete the Ultraman materials involved from its training data, and pay compensation for loss of RMB 300,000, etc. On 1 February 2024, Plaintiff applied to change the cause of action from an internet infringement to a copyright infringement, claiming the infringement of its rights of reproduction, adaptation and information network dissemination by the Defendant.
On 5 January 2024, the case was accepted. Then, it was heard in public by using simplified procedures on 4 February. The Court considers that the key issues in this case are whether the Defendant had infringed the Plaintiff’s rights of reproduction, adaptation and information network dissemination, and if it constitutes an infringement, what kind of civil liability should the Defendant bear?
In judicial practice, the determination of copyright infringement generally requires the satisfaction of two conditions: contact and substantial similarity. The Court held that the Defendant had the possibility of accessing the Plaintiff Ultraman’s work because the work enjoyed a high reputation in China. Meanwhile, it considered the pictures generated by the Defendant’s website are substantially similar to the Plaintiff’s Ultraman’s work because the former partially or completely copied the original expression of the artistic image of the latter, and many key features of the former are extremely similar to the latter. Furthermore, it found the pictures generated by the Defendant’s website partly retained the original expression of the Plaintiff's Ultraman works and formed new features on this basis. The Court held such a creation of the Defendant by AIGC constitutes an adaptation of the Plaintiff's Ultraman works. Therefore, regarding the first key issue, the Court determined that the Ultraman pictures generated by the Defendant’s website infringed the Plaintiff’s rights of reproduction and adaption. As to whether the Defendant infringes the right of information network dissemination, the Court did not evaluate it based on the ground that the infringement was included in the control of the rights of reproduction and adaptation.
Regarding the second key issue, the Court required the Defendant to further prevent users from generating pictures that are substantially similar to the Plaintiff’s Ultraman work when they normally use prompt words related to Ultraman. As for the Plaintiff's request to delete the Ultraman materials from the Defendant’s training data, the Court held that the Defendant was an AIGC service provider rather than a large model developer and had not conducted model training, so it did not support the Plaintiff’s request for deletion. In addition, the Court held that generative AI service providers should take certain technical measures to avoid generating images that are substantially similar to other’s works; and should fulfill their reasonable duty of care, including establishing a complaint and reporting mechanism, alerting potential risks, labeling generative content according to Article 17 of the Provisions on the Administration of Deep Synthesis of Internet-based Information Services issued on 25 November 2022.
34 days later, the Court issued its first-instance judgment, ultimately ruling that the Ultraman pictures generated by the Defendant's website infringed the Plaintiff's right of reproduction and adaption, and ordered the Defendant to stop the infringement and compensate the Plaintiff RMB 10,000 for economic losses, including evidence collection fees, reasonable expenses, etc.
The Judgment on the first AIGC infringement case triggered heated discussions. Some legal experts believe that the Court approximately deduced from the final generated image that AIGC copied the original work. However, the reduction or reproduction of the original image by the AI data model is significantly different from the direct reproduction that has long been prohibited by China’s Copyright Law. Similarly, adaptation in the traditional copyright law refers to re-creation by using the original work in sequential order, but there is no such difference in AIGC technology. Meanwhile, some legal experts question the practicality of deleting specific data from large models. By the safe harbor principle, Internet content platforms only need to perform notice-and-delete obligations for infringing content. However, in this case, the website of the Defendant was still able to generate Ultraman pictures after blocking related keywords. From a technical perspective, deleting specific data in a large model seems to be a complicated problem, because it may not be possible to locate what kind of data the Ultraman picture became after the training is completed.
The judgment of the first AIGC infringement case has sounded the alarm for both ordinary users of AIGC and AIGC service providers. In the process of using AIGC, ordinary users’ behavior of modifying and generating new works based on existing copyrighted works and publishing them to the public may constitute infringement; while AIGC service providers may also be suspected of infringement if they generate pictures of well-known IP without authorization. Compared with ordinary users, the compliance management of AIGC service providers has received more attention from this judgment. Following the Interim Measures for the Management of Generative Artificial Intelligence Services issued on 10 July 2023 and the judgment of this case, AIGC service providers are recommended to establish a convenient complaint and reporting mechanism, to use user agreements and other methods to fully indicate potential risks, and prominently labeling the generated content.
The Ultraman image is a world-renowned IP work and has wide influence and popularity internationally. The copyright holder of Ultraman signed an authorization letter with Shanghai Xinchuanghua Cultural Development Co., Ltd. (“Plaintiff”), exclusively authorizing it the copyright of the Ultraman series of images in China (excluding Hong Kong, Macau, and Taiwan). The Plaintiff found that an AI company (“Defendant”) provided an AI painting service on its website. Member users of the Defendant can use the AI painting service by consuming the computing power purchased by their recharge. Through this service, when a member user requests to generate a picture related to Ultraman, the image in the picture generated by the website is substantially similar to the Ultraman image of The Plaintiff. The works generated by the AI painting service provided by the Defendant are mainly based on large amounts of data training, and the training data are a large number of original works, including the Ultraman work of the Plaintiff. The Plaintiff believed that the Defendant had committed online infringement. Thus, in early January 2024, it filed a lawsuit with the Court, requesting the Defendant to stop generating infringing Ultraman pictures, delete the Ultraman materials involved from its training data, and pay compensation for loss of RMB 300,000, etc. On 1 February 2024, Plaintiff applied to change the cause of action from an internet infringement to a copyright infringement, claiming the infringement of its rights of reproduction, adaptation and information network dissemination by the Defendant.
On 5 January 2024, the case was accepted. Then, it was heard in public by using simplified procedures on 4 February. The Court considers that the key issues in this case are whether the Defendant had infringed the Plaintiff’s rights of reproduction, adaptation and information network dissemination, and if it constitutes an infringement, what kind of civil liability should the Defendant bear?
In judicial practice, the determination of copyright infringement generally requires the satisfaction of two conditions: contact and substantial similarity. The Court held that the Defendant had the possibility of accessing the Plaintiff Ultraman’s work because the work enjoyed a high reputation in China. Meanwhile, it considered the pictures generated by the Defendant’s website are substantially similar to the Plaintiff’s Ultraman’s work because the former partially or completely copied the original expression of the artistic image of the latter, and many key features of the former are extremely similar to the latter. Furthermore, it found the pictures generated by the Defendant’s website partly retained the original expression of the Plaintiff's Ultraman works and formed new features on this basis. The Court held such a creation of the Defendant by AIGC constitutes an adaptation of the Plaintiff's Ultraman works. Therefore, regarding the first key issue, the Court determined that the Ultraman pictures generated by the Defendant’s website infringed the Plaintiff’s rights of reproduction and adaption. As to whether the Defendant infringes the right of information network dissemination, the Court did not evaluate it based on the ground that the infringement was included in the control of the rights of reproduction and adaptation.
Regarding the second key issue, the Court required the Defendant to further prevent users from generating pictures that are substantially similar to the Plaintiff’s Ultraman work when they normally use prompt words related to Ultraman. As for the Plaintiff's request to delete the Ultraman materials from the Defendant’s training data, the Court held that the Defendant was an AIGC service provider rather than a large model developer and had not conducted model training, so it did not support the Plaintiff’s request for deletion. In addition, the Court held that generative AI service providers should take certain technical measures to avoid generating images that are substantially similar to other’s works; and should fulfill their reasonable duty of care, including establishing a complaint and reporting mechanism, alerting potential risks, labeling generative content according to Article 17 of the Provisions on the Administration of Deep Synthesis of Internet-based Information Services issued on 25 November 2022.
34 days later, the Court issued its first-instance judgment, ultimately ruling that the Ultraman pictures generated by the Defendant's website infringed the Plaintiff's right of reproduction and adaption, and ordered the Defendant to stop the infringement and compensate the Plaintiff RMB 10,000 for economic losses, including evidence collection fees, reasonable expenses, etc.
The Judgment on the first AIGC infringement case triggered heated discussions. Some legal experts believe that the Court approximately deduced from the final generated image that AIGC copied the original work. However, the reduction or reproduction of the original image by the AI data model is significantly different from the direct reproduction that has long been prohibited by China’s Copyright Law. Similarly, adaptation in the traditional copyright law refers to re-creation by using the original work in sequential order, but there is no such difference in AIGC technology. Meanwhile, some legal experts question the practicality of deleting specific data from large models. By the safe harbor principle, Internet content platforms only need to perform notice-and-delete obligations for infringing content. However, in this case, the website of the Defendant was still able to generate Ultraman pictures after blocking related keywords. From a technical perspective, deleting specific data in a large model seems to be a complicated problem, because it may not be possible to locate what kind of data the Ultraman picture became after the training is completed.
The judgment of the first AIGC infringement case has sounded the alarm for both ordinary users of AIGC and AIGC service providers. In the process of using AIGC, ordinary users’ behavior of modifying and generating new works based on existing copyrighted works and publishing them to the public may constitute infringement; while AIGC service providers may also be suspected of infringement if they generate pictures of well-known IP without authorization. Compared with ordinary users, the compliance management of AIGC service providers has received more attention from this judgment. Following the Interim Measures for the Management of Generative Artificial Intelligence Services issued on 10 July 2023 and the judgment of this case, AIGC service providers are recommended to establish a convenient complaint and reporting mechanism, to use user agreements and other methods to fully indicate potential risks, and prominently labeling the generated content.