Visual China.
Text: Internet Law Review.On February 8, 2024, the Guangzhou Internet Court made a judgment on the alleged infringement of Ultraman's copyright by the ** generated in the AI services provided to the public by an AI platform, which became the world's first court judgment against the copyright infringement of an AIGC service provider, and received enthusiastic attention from the AI industry and the legal industry. This case is one of the important feedbacks made by the Chinese judiciary as it enters the AI era opened by ChatGPT, and the law is facing many challenges brought about by the development of AI technology.
Among the many AI technical challenges faced by the law, the most concerned is the challenge brought by AI to intellectual property rights, and in AI-related intellectual property cases, there have been three court judgments in China on the copyrightability of AI-generated works, but there has been no court judgment including Chinese courts on how AI service providers should bear legal liability for copyright infringement of AI-generated works, and how AI service providers should bear legal liability. Although there are already a large number of cases in the United States that are in the process of litigation, there has been no effective judgment, and this court judgment has served as a reminder for AI service providers to judge the risk of copyright infringement on their own platforms in the future.
In this case, Tsuburaya Productions, as the copyright owner of the Ultraman series, authorized the plaintiff in this case in 2019 to copyright the image of the Ultraman series. The defendant AI company (pseudonym) operates TAB (pseudonym) **, which has the function of AI-generated painting. The plaintiff found that when the Tab was requested to generate Ultraman-related **, the Ultraman image generated by the Tab was substantially similar to the Ultraman image for which the Plaintiff was copyrighted. The court finally found that TAB**, as an AI service provider, constituted a direct infringement of Ultraman's copyright reproduction rights and adaptation rights, and that it was liable for compensation because the filtering measures taken could not ensure that AI could not generate ** that was substantially similar to Ultraman, and that it failed to fulfill the AI service provider's duty of care.
Since the content generated by the AI platform is generated based on the prompt (prompt word) input by the user to the AI, it has been discussed that even if the AI platform generates infringing content, it is also caused by the user's prompt (prompt word), so the AI platform should be the same as the UGC network platform, and the safe harbor can be applied, that is, the infringement liability can only be borne when the copyright infringement should be known or known.
However, in this case, the judgment of the Guangzhou Internet Court directly determined that the ** generated by the TAB platform that was substantially similar to Ultraman constituted copyright infringement, so in the future, there would be no objection to the AI platform if it generated works that were substantially similar to those of a third-party rights holder.
After the determination of copyright infringement, the law overcomes the legal liability of the infringing party to stop the infringement, and with regard to how the AI platform can stop the infringement, this case is the first to clearly determine that the AI platform can take copyright filtering measures to achieve the effect of stopping the infringement. Before the judgment of this case, all AI platforms were already discussing what measures to take to avoid copyright infringement that they could not control, but which had a very high probability of occurring. This case gives a very good solution, but in this case, it is also determined that "the degree of precaution should be achieved: the user's normal use of prompts related to Ultraman cannot generate ** that is substantially similar to the Ultraman works involved in the case." This seemingly simple requirement,Especially when put into a super IP like Ultraman, it is very difficult to achieve,As many as 50 kinds of Ultraman images make it a very difficult thing for AI not to produce not only not exactly the same as them,Including the ** with similar main features。
Therefore, under the judgment of the court in this case, although the AI platform can take copyright filtering measures if it wants to avoid copyright in the future, how to take copyright filtering measures that can prevent the generation of substantially similar content will further test the compliance and technical capabilities of the AI platform.
Since the determination of copyright infringement needs to be based on whether the infringing party has subjective intent, this case is a pioneering determination that the AI platform's duty of care is determined as whether it constitutes the subjective intent of copyright infringement, so that the AI platform can be exempted from liability for compensation when it fulfills its duty of care. The Guangzhou Internet Court held that, in accordance with the Interim Measures for the Administration of Generative AI Services and the Provisions on the Administration of Deep Synthesis of Internet Information Services, AI service providers should take actions such as establishing a reporting mechanism, alerting potential risks, and making conspicuous markings, and based on this, determine the subjective intent of the AI platform and determine whether it should bear infringement liability.
This solution can undoubtedly greatly reduce the risk of copyright infringement for AI platforms, but whether this judgment can be recognized by more courts still needs to wait for further determination by higher-level courts in other cases.
At the same time, from the perspective of practical implementation, even if the AI platform can judge whether to perform the duty of care in accordance with the Interim Measures for the Administration of Generative AI Services, so as to avoid the liability for copyright infringement, it is still difficult for the platform to take measures to meet the requirements that the keyword filter does not produce the effect of constituting a substantive right, so it is difficult to avoid the liability to stop the infringement.
In this case, the TAB platform, as an AI service provider, did not develop its own base model, that is, it did not directly use Ultraman and other images for model training, but as an AI service provider, although the TAB platform claimed that it did not use Ultraman's works for training, it still bears copyright infringement liability. In this case, the court did not explain the issue of the allocation of tort liability between the base model provider (the party that directly conducts model training) and the AI service provider. This remains to be determined in other relevant cases.
The judgment of this case shows that with the advent of the AI era, the judiciary will face subversive determinations of new technologies, and the responsibilities of new industry participants arising from new technologies will also become more and more detailed with the increasing number of judicial cases.