An Exploration of Copyright Protection for AI-Generated Outputs

I. Overview of AI-Generated Outputs
John McCarthy, the father of artificial intelligence and a Turing Award laureate, once observed that "artificial intelligence is about making the behavior of machines look like the intelligent behavior exhibited by humans." The essence of generative artificial intelligence (generative AI) lies in learning the patterns and features of large volumes of data and conducting deep learning and pattern recognition so as to generate content that is novel and creative. Its operating logic follows a cyclical process of "preparation—computation—generation": in the preparation stage, model learning and data preparation are carried out; in the computation stage, user feedback data is introduced, data is preprocessed, features are extracted, and the model is trained; and in the generation stage, content that is similar to the original data yet novel and distinctive is output, which in turn affects society. The greatest characteristic of generative AI lies in the very nature of "generation." It can essentially independently complete the extrapolation and updating of data and devise response strategies for complex and volatile problems. Unlike ordinary AI, which depends on technicians to update its data, generative AI can spontaneously complete updates and iterations; indeed, it is like a "black box," such that even technicians cannot fully comprehend the computational processes within it.
AI-generated outputs refer to the type of creation that is autonomously generated by artificial intelligence without any human intervention; even where human intervention exists in the content creation process, it is confined to activities at the preliminary production stage, such as writing source code, "feeding" the algorithm, and setting objectives. Depending on the generation mode, such outputs can be classified into text-to-text, text-to-image, and image-to-image modes. Generative AI is significantly stronger than analytical AI in its capacity to process and analyze data; its workflow for generating content includes automated compilation of content, intelligent polishing and refinement, multimodal conversion, and creative generation—thereby giving rise to the core question of whether the outputs of generative artificial intelligence are protected by copyright.
II. The Copyrightability of AI-Generated Outputs
Whether AI-generated outputs can constitute "works" within the meaning of copyright law—that is, the question of copyrightability—is the primary issue in the copyright protection of artificial intelligence. Judicial practice has taken differing attitudes toward this question and has produced a number of landmark cases.
A. Cases in Which AI-Generated Outputs Constitute Works
Case 1 — Beijing Internet Court (2023): China's first "AI text-to-image copyright infringement" case
In China's first "AI text-to-image copyright infringement" case, heard by the Beijing Internet Court in 2023, the plaintiff used the open-source software Stable Diffusion to generate the image at issue, "The Spring Breeze Brings Tenderness," by entering prompts, and published it on the Xiaohongshu platform. The defendant's use of the image without authorization and its cropping of the attribution watermark gave rise to the infringement dispute. After trial, the court held that the image satisfied the definition of a work and constituted a work. As to the appearance of the image, it had been generated by the plaintiff using generative AI technology; as to the creation process, the plaintiff had made a certain intellectual input from conception through to the final selection of the image—designing how the characters were presented, selecting prompts, arranging the sequence of the prompts, setting relevant parameters, and choosing the final image—and the image therefore satisfied the "intellectual achievement" requirement. As to originality, the image exhibited identifiable differences from prior works, and the plaintiff had designed and arranged the pictorial elements and the compositional layout through prompt and parameter settings; this process of adjustment and revision reflected the plaintiff's aesthetic choices and individual judgment, satisfying the "originality" requirement. The court ultimately held that the image at issue was a work of fine art and was protected by copyright law.
Case 2 — Beijing High People's Court (2026): the "virtual digital human infringement" case
In the "virtual digital human infringement" case released by the Beijing High People's Court in 2026, the image of the virtual digital human "Tianyu" was held to constitute a work of fine art. The court held that the image of the virtual digital human did not derive directly from a real person but was created by a production team, and that it reflected distinctive aesthetic choices and judgment in the design of lines, colors, and the specific image, thereby meeting the originality requirement for works and deserving copyright protection. The case shows that where the creation process reflects a sufficient intellectual input and individualized expression, the resulting output may still obtain copyright protection even though AI tools have been used.
B. Cases in Which AI-Generated Outputs Do Not Constitute Works
Case 1 — Zhangjiagang City People's Court (first instance) and Suzhou Intermediate People's Court (appeal): the first case holding that "AI text-to-image output does not constitute a work"
In the first case holding that "AI text-to-image output does not constitute a work," heard by the Zhangjiagang City People's Court, the plaintiff, Feng, created the "Phantom Wings Transparent Art Chair" series of images using an AI text-to-image application. The court held that the user should provide the original records of the creation process to prove that, by adding prompts and modifying parameters, he had adjusted, selected, and refined the initially generated images, thereby showing that he had made individualized choices and substantive contributions with respect to the expressive elements of the images—such as layout, proportion, perspective, compositional elements, color, and lines. Because the plaintiff failed to provide the corresponding original records, such as workflow diagrams, from the creation process, and admitted that, given the randomness and uncertainty of the software's image generation, the process that produced the same content could no longer be reproduced, the court found it difficult to determine that he had made specific individualized choices and substantive contributions, and accordingly held that the images at issue did not satisfy the constituent elements of a work. In April 2025, the Suzhou Intermediate People's Court ruled on appeal to uphold the original judgment.
Case 2 — Huangpu District People's Court of Shanghai: China's first "AI painting prompt copyright dispute"
In the case heard by the Huangpu District People's Court of Shanghai—China's first "AI painting prompt copyright dispute"—a cultural company in Shanghai generated paintings on an AI platform by writing a series of prompts, and later discovered that users Zhu and Sheng had used identical prompts to generate highly similar paintings and had published them publicly. The company accordingly brought a copyright infringement action. After trial, the court held that prompts constitute abstract creative ideas and, in both content and form, do not qualify as "works" within the meaning of copyright law; that China's Copyright Law protects "expression that is original" rather than abstract ideas; and that, in this case, the elements of the prompts were merely listed in simple juxtaposition and constituted conventional expression in the field. The court therefore dismissed all of the plaintiff's claims. This case is China's first judicial decision to focus on the copyright status of AI prompts, and it clarifies the boundary of rights between users of AI creation tools and AI-generated works.
Case 3 — Yuehu District People's Court of Yingtan City, Jiangxi Province: the "one-click AI image generation" case
In the "one-click AI image generation" case heard by the Yuehu District People's Court of Yingtan City, Jiangxi Province, Cui entered text describing natural scenery into an image generation application and, with a single click, generated four landscape images. Cui later discovered that a cultural and creative company had used the image on its WeChat official account. The court held that the AI software at issue generated images automatically upon the mere entry of text prompts, and that the results generated from identical keywords were random each time and lay beyond the user's ideational control; Cui had merely entered solar-term and scene descriptions without any manual modification or parameter adjustment, and the creation process relied heavily on technical functions and involved no intellectual labor commensurate with that of traditional creation. The images at issue therefore lacked the author's original expression and did not constitute original works protected by copyright law. The case was selected as one of the "Top Ten Cases Advancing the Rule of Law in the New Era 2025."
Case 4 — Chengdu Intermediate People's Court: the "AI-generated image copyright infringement dispute"
In the "AI-generated image copyright infringement dispute" heard by the Chengdu Intermediate People's Court, Cao, a digital art creator, entered keywords such as "China, guofeng (traditional Chinese style), festival, lanterns" into the "Jimeng AI" application and, after repeated adjustments, generated an artwork image, which he then registered for copyright. He subsequently discovered that a tourism college had used the image on its WeChat official account. The court held that copyright law protects intellectual achievements that are original, and that in Cao's process of generating images with artificial intelligence his contribution consisted principally of supplying basic, conventional prompt terms, without substantive intellectual input sufficient to reflect the creator's personality and creativity in the formation of the final image; the generated content therefore did not satisfy the originality requirement for works. The case established the "intellectual input and expression of personality" review standard, providing operable adjudicative guidance for similar disputes.
C. Trends in Determining the Copyrightability of AI-Generated Outputs
Under Article 3 of the Copyright Law of the People's Republic of China, a "work" means an intellectual achievement in the fields of literature, art, and science that is original and capable of being expressed in a certain form. The core requirements for copyrightability comprise two aspects: "originality" and "intellectual achievement." Originality requires that a work be completed by the author through independent conception and the investment of creative intellectual labor, rather than by copying the work of others. The "intellectual achievement" requirement calls for determining whether the object in question was produced through mechanical intellectual labor, skilled intellectual labor, or creative intellectual labor; only the last of these is eligible for copyright protection.
In the era of generative AI, the greatest challenges facing copyright law are the originality of the "work" (which is machine-generated) and the subjectivity of the "author" (whose identity is that of a machine). Scholars hold divergent views on this point. Opponents of protection argue that AI-generated content is merely the result of the operation of algorithms, rules, and templates, lacking the mental and conscious creation contemplated by copyright law, and should therefore fall within the public domain. Proponents of protection argue that AI works possess objective value approaching that of traditional works and are consistent with copyright law's fundamental objective of encouraging creation, and that consigning AI-generated content to the public domain could squeeze the market share of traditional works; copyright protection is therefore a practical necessity.
III. Ownership of Copyright in AI-Generated Outputs
Once the copyrightability of AI-generated outputs is established, the core question that follows is: to whom should the copyright belong? The traditional logical structures of "author–work" and "subject–object" face severe challenges in the AI era.
A. Judicial Practice: the "Filin v. Baidu" Case
In the copyright infringement dispute brought by Beijing Filin Law Firm against Baidu, the firm used the "Weike Xianxing" legal database to generate a legal big-data analysis report, on the basis of which it compiled a complete analysis report. The defendant published the report after removing the attribution without authorization, giving rise to the infringement dispute. The court held that, in the absence of contrary evidence, the subject of creation of a work should be recognized as the subject of rights, and that the article at issue was a work of a legal person created by Filin Law Firm. The case established the basic rule that "the subject of creation is the subject of rights," while exercising a cautious approach with respect to the degree of AI involvement in creation.
B. Determining the Ownership of Copyright in AI-Generated Outputs
The emergence of generative AI has posed profound challenges to the traditional "anthropocentric" theory of copyright ownership, and scholars have proposed a range of theoretical approaches:
First, the theory of virtual legal personality. This theory holds that AI possesses a certain autonomy and independence and should be granted a certain legal status; however, current AI systems have no "autonomous consciousness," and granting AI legal personality would give rise to moral and ethical problems.
Second, the designer theory. This theory holds that the algorithms, programs, and training data behind AI are supplied by human designers, and that rights should therefore belong to the designers.
Third, the user theory. This theory holds that the users who actually activate and apply the AI have a direct influence on its outputs, and that users should enjoy the rights.
Fourth, the joint theory. This theory maintains that the "designers" and "users" of AI hold the rights in the work jointly.
Fifth, the public domain theory. This theory holds that works created by AI should be placed in the public domain for free use.
Each of these theories has its own theoretical foundation, yet in judicial practice the tendency is to adopt the position of the "user theory," applying the criterion of "necessary human intervention in the work"—that is, whether the human user's contribution reaches a level that determines the essential attributes of the work. Where a user has made substantive contributions to the generated content through prompt design, parameter adjustment, content selection, and similar means, that user should be recognized as the author.
Conclusion
The rapid development of generative artificial intelligence technology is profoundly reshaping the ecosystem of content creation. In the face of this transformation, copyright law—as the core legal institution for protecting and incentivizing creation—has, while upholding the fundamental position of "human-author centrism," also demonstrated the necessary institutional flexibility. Judicial practice has made useful explorations in relevant cases, seeking an appropriate balance between protection and innovation, building judicial rules for copyright protection suited to the AI era, and realizing the value leadership of the copyright system in the age of artificial intelligence.