Copyright infringement in content generated by AI: an empirical analysis based on typical cases of China and the United States

The mathematical model for factor analysis is as follows: (1)yk=uk+ak1f1+ak2f2+⋯+akmfm+εk,k=1,2,⋯,n

The general model is: (2){Y1=μ1+a11F1+a12F2+⋯+a1mFm+ε1Y2=μ2+a21F1+a22F2+⋯+a2mFm+ε2⋮Yρ=μρ+aρ1F1+aρ2F2+⋯+aρmFm+ερ

In equations (1) and (2), y=(Y1,Y2,⋯,Y3) is a n-dimensional random variable consisting of n indicator elements, f=(F1,F2,⋯,F3) is a common factor that can represent and explain the original variable. ε_k is the random variable, a_k is the factor loadings, and the larger a_k the correlation, the stronger the correlation. ∑i=1maki is the variance contribution ratio, which is a numerical representation of the extent to which the extracted public factors contribute to the y_k score of each public factor F_K: (3)fj=βj1y1+βj2y2+⋯+βjmym,j=(1,2,⋯,m)

In equation (3), f_j is the score of the jnd common factor, β_j1, β_j2, ⋯, β_jm is the score coefficient of the factor, and y₁, y₂, ⋯, y_m is the standardized variable. The ratio of the variance contribution of each public factor to the cumulative variance contribution is regarded as the weight coefficient of the public factor, and then the scores of each sample on each public factor are weighted and summarized to obtain the composite score.

The KMO test examines variable data based on the relationship between the simple and partial correlation coefficients between variables. When the sum of the squares of the simple correlation coefficients of all variables is much larger than the sum of the squares of the partial correlation coefficients, the stronger the correlation between the variables, the more suitable for principal component analysis. Let (Xi,Yi)(i=1,2,3,…,n) be a sample taken from the overall population, then the Pearson simple linear correlation coefficient of the sample is calculated as: (4)ρ=∑i=1n(Xi−X¯)(Yi−Y¯)∑i=1n(Xi−X¯)2∑i=1n(Yi−Y¯)2 (5)X¯=1n∑i=1nXi,Y¯=1n∑inYi

The partial correlation coefficient is a calculation of the magnitude of the correlation coefficient between two variables under the influence of a fixed number of remaining variables, reflecting the degree of linear correlation between any two variables under the influence of a fixed number of remaining variables. The calculation of partial correlation coefficient is shown in equation (6) and equation (7): (6)rxy,z1=rxy−rxz1ryz1(1−rxz12)(1−ryz12)(h=1) (7)rxy,z122z3⋯zh=rxy,z1z2z3⋯zh−1−rxzh,z1z2z3⋯zh−1ryzh,z1z2z3⋯zh−1(1−rxzk2,z1z2z3⋯zh−1)(1−ryzh2,z1z2z3⋯zh−1)(h≥2)

In equations (6) and (7): h is the number of variables fixed. r_xy is the simple correlation coefficient between variables x and y. Z₁, Z₂, …, Z_n is the fixed variable. Let P be the sum of squares of the simple correlation coefficients of the variables and R be the sum of squares of the partial correlation coefficients, and the KMO test statistic is calculated as: (8)M=PP+R

Before conducting the factor analysis, we used SPSS software and Python programming to conduct a KMO test for each set of categories; if the KMO value is greater than 0.5, there is a strong correlation between the variables and the data can be considered suitable for factor analysis.

On the basis of the refinement of the key elements of the intellectual property system, in order to further explore the impact of AI creations on the key elements of intellectual property subject, object and content, this study adopts a questionnaire survey to carry out relevant research. The survey objects of this study are AI R&D/technical personnel, AI product personnel, intellectual property law and policy researchers, intellectual property intermediary service personnel, etc. 794 questionnaires were distributed, and 761 questionnaires were recovered, with a questionnaire recovery rate of 95.84%. There are 750 valid questionnaires, which provide detailed and rich data support for the subsequent exploratory factor analysis and validation factor analysis.

Descriptive statistical information is obtained as shown in Table 1. It shows that AI intellectual property-related practitioners have a richer knowledge reserve, which is in line with the academic distribution structure of workers at the forefront of science and technology such as artificial intelligence/intellectual property. In terms of occupation, AI R&D personnel accounted for 35.73%, IP law and policy researchers accounted for 6.13%, and IP service personnel accounted for 8.4%.SPSS statistical software was used to conduct the reliability test of the scale, and the statistical results showed that the overall Cronbach’s alpha coefficient value was 0.898, which was greater than 0.7, and the CITC values were all greater than 0.4, indicating that the overall reliability of the scale is good.

With the help of SPSS statistical software, the study took the lead in applying the principal component analysis method to extract the male factors, and the setting of the male factor extraction method was set to not fix the number of factors. The factors with eigenvalues greater than 1 were extracted for factor rotation, and after iteration, the results were consistent with the results of the pre-study, as shown in Table 2. A total of seven factors were extracted, with a cumulative variance contribution rate of 63.721%.

Among them, Factor 1 focuses on the issues of AI creation author judgment, rights attribution, and liability, so Factor 1 is named the attribution of rights of AI creation. Factor 2 reflects the possibility of adjusting the standard for judging the originality of AI, thus naming Factor 2 as the judgment of originality of AI creations. Factor 3 discusses the term of protection of AI creations and the content of copyright protection, thus naming it as special provisions for copyright protection of AI creations. Factor 4 focuses on the necessity and feasibility of AI creations becoming the object of intellectual property rights, thus named as the judgment of copyrightability of AI creations. Factor 5 focuses on the influence of AI creations on the infringement judgment and infringement proof process of the patent system, which can be named as the mechanisms for dispute resolution. Factor 6 focuses on the conceptual connotation and characteristics of AIGC, which can be named as AI creations. Factor 7 has a larger loading on variables such as the risk of infringement of artificial intelligence creations, the determination of copyright infringement, and the legal regulation of infringement, so factor 7 can be named as the determination of copyright infringement of artificial intelligence creations.The total of seven common factors obtained from the EFA results indicate that the questionnaire has a more reasonable structural validity.

Based on the conclusions drawn from the exploratory factor analysis, this research study followed up with a validation factor analysis (CFA) to better utilize structural equation modeling to further determine the aspects of the impact of AI creations on the intellectual property system, and the maximum likelihood method was used as the parameter estimation method to obtain standardized factor loadings for each of the items, and the results showed that each of the standardized factor loadings ranged between 0.53 -0.89. The analysis indicates that among the seven identified factors, the judgment of copyrightability, the judgment of originality, the attribution of rights, and the mechanisms for dispute resolution, are statistically the most significant.

There are various test criteria for CFA, and root mean square of approximation error (RMSEA) and goodness-of-fit index (GFI) are all important indicators for evaluating the goodness-of-fit. In this study, the following six commonly used fitting metrics are selected as test criteria as shown in Table 3. All six indicators show that the overall goodness of fit of the model of the impact of AI creations on intellectual property is no less than 0.8, and the overall effect is good.

Through factor analysis of the sample data, this paper finds that based on the technical characteristics of automation, adaptability, information interactivity, non-predictability of results, evolvability, and optimization of goal orientation, GenAI will carry out acts such as literary and artistic creation, and it is difficult to differentiate the outward expression of the creation results from the human works, which directly brings great difficulties to the delimitation of the protection scope of the intellectual property system. Statistically, the judgments regarding copyrightability and originality, the attribution of rights, and the mechanisms for dispute resolution emerge as the most critical aspects. The subsequent sections will analyze specific cases to examine how courts in China and the United States assess these key factors. However, given the current scarcity of case law—which primarily centers on copyrightability issues—the analysis below will focus chiefly on this aspect.

As GenAI continues to enhance its capacity to replicate human creativity, it is only a matter of time before it surpasses human authorship, leading to an increase in related disputes. For example, in China, several landmark cases have played a pivotal role in shaping the legal framework surrounding AIGC and IPR. Notably, the Li v. Liu case, adjudicated by the Beijing Internet Court, addressed the infringement of the right to disseminate information online [(Beijing Internet Court, Li v. Liu, Case No. (2023) Jing 0491 Min Chu 11279)]. Similarly, Lin Chen v. Hangzhou Gaosi Qimai Technology Co., Ltd. and Changshu Qinhong Real Estate Development Co., Ltd., ruled upon in 2024 by the People’s Court of Changshu City in Jiangsu Province, further contributed to the evolving legal discourse on AIGC and IPR[(Civil Judgment of the People’s Court of Changshu City, Jiangsu Province, Case No. (2024) Su 0581 Min Chu 6697)].

At the same time, in the United States, numerous legal disputes have also emerged concerning AIGC. Cases such as Sarah Andersen v. Stability AI Ltd. and Getty Images (US), Inc. v. Stability AI Ltd. exemplify ongoing litigation over AI-generated works. More recently, in February 2025, a highly anticipated ruling was issued in Thomson Reuters v. Ross Intelligence, marking what is likely the first U.S. judicial decision addressing AI training and copyright infringement (though it does not specifically pertain to GenAI). In this case, the U.S. District Court of Delaware conducted a comprehensive analysis of whether the use of copyrighted materials in AI training constitutes fair use under U.S. copyright law—a topic of global significance.

However, due to the automated nature of AIGC, traditional methods of determining copyright infringement are not entirely applicable. Consequently, judicial interpretations vary significantly across jurisdictions, and a universally accepted legal framework has yet to be established. Examining the judicial decisions of China and the United States, several landmark cases related to AIGC stand out as particularly noteworthy. These cases not only represent significant milestones in the legal discourse on AIGC but also warrant a detailed comparative analysis to assess their broader implications for the evolving legal framework surrounding AIGC.

Throughout the discourse on this subject, the “monkey selfie” case has ignited particularly fervent debate and exerted a considerable influence on disputes pertaining to AIGC. The Monkey Selfie case occurred in 2011, when a monkey took a picture with photographer Slater David’s camera. David argued that he had a copyright on the photo. The U.S. Copyright Office(USCO) issued a document in 2014, treating the monkey selfie as a classic case, stating that the work was actually a work created by a human being, thereby claiming that the actions of animals are not protected by copyright law.In 2015, the Animal Rights Defense Institute filed a lawsuit demanding that all proceeds received from the photo must be used for the monkey and its habitat, arguing that the copyright owner of the selfie was the monkey. The court dismissed the suit on the grounds that works created by animals are not protected by copyright. Thus, even if a work meets the formal rules of copyrightability, the absence of human operation—such as a human pressing the shutter—and a lack of human creativity preclude its protection under copyright law. In such cases, the USCO would refuse registration for works produced solely by automatic or mechanical processes without the requisite level of human control or creative input.

For this reason, in early 2022, USCO denied copyright protection for Stephen Thaler’s AI-generated artwork- “A Recent Entrance to Paradise”. The rejection was based on the principle that a work must be created by a human author to qualify for copyright protection. The USCO emphasized that works produced solely by machines or through purely mechanical processes, without any creative input or intervention from a human author, are ineligible for copyright registration. The USCO also revoked the copyright registration of Zarya of the Dawn, a graphic novel created by Kris Kashtanova using synthetic media technology. The work, generated with the assistance of Midjourney, included AI-created characters, dialogue, and other elements.

Among judicial rulings in China, the most representative case involving AI works is the DreamWriter case.In August 2018, the plaintiff, Shenzhen Tencent Co., Ltd. (hereinafter referred to as “Tencent”), filed a lawsuit with the Nanshan District Court on the grounds that the defendant Shanghai Yingke Technology Co., Ltd. infringed the article in question completed by it using DreamWriter software. The Court held that the article in question met the requirements of the external form of a written work, and that the intervals that existed in the creative process were only due to the tools used, while in terms of the specific arrangements for originality, the original team had a direct link of intellectual activity with the particular form of expression of the article in question. The combination of its external form and the “personalized arrangement” of the creative process should have originality. And based on the fact that the article in question was expressed by the plaintiff’s team as a whole, it constituted a legal person’s work and the original team could sign its name.

This case is the first case of determining the copyright protection of AIGC, which mainly focuses on the object and ownership of DreamWriter’s creative content, and its main focus is as follows. First, the judgment of objective originality. The court held that the external expression of the article in question embodied the selection and arrangement of information and data related to the stock market, and that the structure of the article as well as the logic of expression were reasonable and clear, thus meeting the requirement of originality.Secondly, the generation process embodies “personalized arrangement”. The Court held that, firstly, the generation process of the article in question mainly went through four stages, namely, data service, triggered writing, intelligent calibration and intelligent distribution. These four links are the relevant arrangements of the plaintiff’s team, only in the creative process because of the characteristics of the technology led to a lack of synchronization of creative behavior, so it should be found that the plaintiff’s team of relevant personnel involved in the creative process. Second, the article in question was based on the “personalized arrangement” of the creative process by the plaintiff’s team, and was an intellectual activity with a direct link. Therefore, the article in question possessed the elements of a “work”.Third, the work in question constituted a legal person’s work. The Court held that the article in question, as a whole, reflected the needs and intentions of the plaintiff’s team for stock review articles, and satisfied the substantive elements of Article 11 of the Copyright Law(2010 Amendment) of China regarding works of legal persons, as well as the formal elements of attribution.

As shown in Figure 1, the guiding significance of this case for the copyright protection of AIGC lies in the particularity of the creation process of DreamWriter as a tool and technology required for the human creative process, and the “personalized arrangement” of human participation is reflected in the overall process, based on which it can be copyrighted as a work of a legal person. However, there are also shortcomings in this case, which did not prove the independent completion of originality.

Figure 1.

The DreamWriter case court judgement mind map

Based on similar considerations, in December 2023, the Beijing Internet Court ruled in China’s first AI-generated image case (Li v. Liu, concerning the infringement of the right to disseminate information online) that AI-generated images possess a certain degree of originality. Consequently, the court held that copyright ownership should belong to the user who employed the AI tool.

Similarly, in the 2024 in the case of Lin Chen v. Hangzhou Gaosi Qimai Technology Co., Ltd. and Changshu Qinhong Real Estate Development Co., Ltd., , the first-instance court found that the plaintiff had used Midjourney and Photoshop software, modifying prompts, iterating images, and making personalized alterations and choices in the final expression. Given the complex post-processing involved, the court ruled that these efforts sufficiently demonstrated the plaintiff’s creative contribution to the originality of the images in question, thereby affirming their eligibility for copyright protection.

Intellectual property cases from both China and the United States share a common characteristic: the final works in question were not directly created by human authors. However, while the U.S. Monkey Selfie case, for example, is often cited in discussions on non-human authorship, it fundamentally differs from the context of AIGC. Unlike AI, which inherently involves human intervention from developers, designers, and deployers, the Monkey Selfie case lacked any form of human contribution in the creative process. Despite this distinction, all relevant cases share a key similarity—namely, that the final output was not fully controlled by a human user. These cases serve as highly representative examples within the legal discourse on AI and IPR, offering valuable insights for future legal studies in this evolving field.

Courts in both China and the United States have provided detailed interpretations of these cases, and the fundamental components of copyrightability remain largely consistent across these two jurisdictions. As a result, these rulings serve as well-developed and valuable precedents for reference. However, notable distinctions exist in their adjudication approaches. A comparative analysis of these cases reveals that while courts in both countries adhered to their respective copyright legal frameworks and fundamental principles, they placed emphasis on different considerations. Besides, the courts adopted different adjudication logic for individual core issues, which resulted in inconsistent standards for determining the copyrightability of AIGC. A comparison of judicial decisions reveals that courts in both China and the United States primarily consider interest relations, alongside deeper economic and political factors. The following sections provide a detailed discussion of these aspects. To gain a comprehensive understanding of the issues explored in this study, it is essential to adopt a dual perspective integrating law and economics. A thorough examination of the underlying economic factors influencing judicial decisions is crucial for gaining deeper insights into the rationale behind the granting or denial of IPR. 1)

Decision-making on U.S. Intellectual Property Protection Policy

Promote IPR is an important strategy that has been developed since the 1900s under the joint impetus of the U.S. Congress, the executive branch led by the President, the courts, and the interest groups, and has been continuously strengthened since the beginning of this century. The following first analyzes and reviews the process of introducing the U.S. strategy for strengthening international protection of AI intellectual property rights in the 2000s. Then it analyzes the specific policy system implemented by the United States to realize the strategy, especially the specific policy system of the United States to strengthen the international protection of IPR since the beginning of this century and since the financial crisis of 2007. Because these policies are also the policies that have the greatest impact on China’s present and future AI IPR protection.

As the hegemonic position of the United States established after World War II has gradually declined since the turn of the century, the U.S. international trade system has gradually become more defensive and has implemented stricter IPR protection as a means of improving its deteriorating trade balance and preventing a further decline in its competitive advantage. U.S. global trade in goods statistics shown in Figure 2, the United States from the year began to deficit throughout the 80’s or 90’s, every other year there is a deficit and the size of the deficit is expanding.

Figure 2.

American goods trade profile

The United States General Accounting Office (GAO) report “Quantifying the Economic Impact of Piracy and Counterfeiting” states that counterfeiting and piracy affect consumer behavior and the incentives for businesses to innovate. In fact, the International Intellectual Property Alliance and the copyright industry it represents are most concerned with the sector’s economic interests, and have specifically calculated foreign piracy rates and annual losses due to piracy, the results of which are shown in Table 4.

Although there are some problems with its methodology for calculating piracy losses, and the source of its data, it conveys an important message: piracy leads to significant losses for the U.S. copyright industry, with piracy costing the U.S. as much as $4,425,000,000 in 2023. The U.S. must be adamant in demanding and monitoring stronger copyright protection from its important trading partners. This is also a direct reflection of the direct economic interest in promoting the global protection and enforcement of AI intellectual property rights. Therefore, it must take into account the interests of the public, particularly those of major corporations. For example, in the Thomson Reuters v. Ross case, the judge explicitly stated that the primary consideration is whether AI training substantially harms the market value of the original work. Additionally, the transformative nature of fair use must demonstrate an added public benefit.

In summary, the system of measures for strengthening AI IPR in the United States is outlined in Table 5. The enhancement of IPR protection in the AI era is achieved through a comprehensive policy network, which allows for the flexible selection and combination of various instruments. All these policy options are closely linked to market access in international trade, bolstered by the U.S’ formidable scientific, technological, and economic strengths, and driven by robust political power. Notably, the close cooperation and strong alignment between the United States Congress and the Executive Branch regarding GenAI intellectual property interests constitute a fundamental source of this influence. Furthermore, these policies are propelled by substantial political and economic forces. In formulating and implementing such measures, the U.S. government must navigate a complex landscape of interests, carefully balancing public sentiment, international law requirement and with the demands of major corporations. Table 5.

Measures to strengthen IPR in the US