The Beijing Internet Court Examines Originality in AI Copyright Case
Published 19 September 2025
Sarah Xuan
On September 16, 2025, the Beijing Internet Court released a representative case concerning copyright disputes over artificial intelligence-generated content (AIGC).
The case makes it clear that if a right holder claims copyright in an AI-generated image but cannot provide complete evidence of the generation process to prove their original intellectual input, the work does not constitute a “work” under the Copyright Law, and their claim for rights protection will be dismissed.
The plaintiff, Zhou, is a content creator in the cultural and creative industry. Zhou claimed that during a start-up collaboration with the defendant, a Beijing technology company, Zhou independently created an image titled “Cat Crystal Pendant” using AI drawing software and published it in a WeChat group. Later, without reaching a usage agreement, the defendant used the image for commercial promotion without authorization. After Zhou demanded removal, the defendant deleted it. However, in March 2024, Zhou found that the image was used again for promotion, and thus filed a lawsuit, claiming infringement of the right of authorship and the right of information network dissemination, and sought damages and a public apology.
The defendant argued that the disputed image was not created solely by Zhou, claiming the two parties had jointly conceived the AI prompts, material settings, and other elements; that Zhou could not prove the creative process; that the image lacked originality; and that the image was unrelated to actual products for sale and had no profit-making purpose, so it did not constitute infringement.
Originality Determination Is Key The court pointed out that this case is a copyright infringement dispute, and the first step is to determine whether the disputed image constitutes a “work” as defined in Article 3 of the Copyright Law. According to the law, a work must meet four requirements simultaneously: 1. It must belong to the field of literature, art, or science;2. It must be original;3. It must be expressed in a certain form; and4. It must be an intellectual achievement.
In this case, the disputed image is visually similar to common paintings or photographs, belonging to the artistic field and having an external form of expression, thus meeting requirements (1) and (3). However, whether it reflects original intellectual achievement is the core issue in dispute.
Burden of Proof for AI-Generated Content
The court cited Article 3 of the Regulations for the Implementation of the Copyright Law, emphasizing that creation is the intellectual activity that directly produces a work.Compared with traditional painting, the process of generating AI works is more multilayered and indirect. Therefore, the party claiming rights must explain their specific creative ideas, input prompts, selection and modification of the generated content, and submit evidence showing their original intellectual labor, such as prompts, iteration records, and modification records.
This requirement essentially applies the general principle of “he who asserts must prove” to the field of AI-generated works, ensuring that only results with genuine original intellectual labor can be protected under the Copyright Law.
Insufficient Evidence from the Plaintiff
The court held that the plaintiff’s evidence was clearly insufficient: 1. They failed to submit the original process records showing how the disputed image was generated in the AI software.2. The provided “describe” result was merely a post hoc description and could not reproduce the prompts used during generation.3. The so-called “reproduction description” was only a simulated attempt afterward, and the software/hardware environment, operation commands, and steps differed from those at the original time, so it could not prove that Zhou invested original intellectual labor during the initial generation.
Meanwhile, the reproduced result differed in style and composition from the disputed image, making it difficult to prove identity of the generation process. Overall, the current evidence was insufficient to determine that the disputed image had originality, and therefore it did not constitute a “work” under the Copyright Law.
The court of first instance dismissed all of Zhou’s claims. Zhou appealed, but the court of second instance upheld the original judgment, which has now taken legal effect.
The judge pointed out that although the number of intellectual property cases involving AI and big data is currently small, they are growing rapidly. For determining the originality of AI-generated works, the general principle of “he who asserts must prove” should be adhered to. Since AI generation requires lower intellectual input from natural persons compared to traditional painting, determining its originality must depend on case-specific evidence. Creators can protect their rights in the following ways: 1. When using AI to generate content, creators should retain complete generation records such as prompts, iteration processes, selection and modification records, and drafts to prove their intellectual labor and personalized expression.2. In addition to prompts, they may also preserve version comparisons, generation logs, screen recordings, and software parameter settings as objective evidence for use in ownership disputes.
In addition, industry actors should enhance the traceability capabilities of AI models and cooperate with institutional frameworks such as the Administrative Measures on Identifying AI-Generated and Synthesized Content, to prevent abuse of the copyright system and protect genuine innovation.
Comment
This case is one of the first domestic precedents to clearly establish standards for determining copyright in AI-generated images. Its judicial reasoning serves as a reference for future intellectual property protection in the AIGC field. It reminds creators that recognition of copyright does not depend solely on the finished product, but also on whether the “intellectual input” in the creative process can be proven. In the context of rapid AIGC development, building an evidence-conscious mindset of “leaving traces of the process” will be a key prerequisite for claiming copyright.
The defendant argued that the disputed image was not created solely by Zhou, claiming the two parties had jointly conceived the AI prompts, material settings, and other elements; that Zhou could not prove the creative process; that the image lacked originality; and that the image was unrelated to actual products for sale and had no profit-making purpose, so it did not constitute infringement.
Originality Determination Is Key The court pointed out that this case is a copyright infringement dispute, and the first step is to determine whether the disputed image constitutes a “work” as defined in Article 3 of the Copyright Law. According to the law, a work must meet four requirements simultaneously: 1. It must belong to the field of literature, art, or science;2. It must be original;3. It must be expressed in a certain form; and4. It must be an intellectual achievement.
In this case, the disputed image is visually similar to common paintings or photographs, belonging to the artistic field and having an external form of expression, thus meeting requirements (1) and (3). However, whether it reflects original intellectual achievement is the core issue in dispute.
Burden of Proof for AI-Generated Content
The court cited Article 3 of the Regulations for the Implementation of the Copyright Law, emphasizing that creation is the intellectual activity that directly produces a work.Compared with traditional painting, the process of generating AI works is more multilayered and indirect. Therefore, the party claiming rights must explain their specific creative ideas, input prompts, selection and modification of the generated content, and submit evidence showing their original intellectual labor, such as prompts, iteration records, and modification records.
This requirement essentially applies the general principle of “he who asserts must prove” to the field of AI-generated works, ensuring that only results with genuine original intellectual labor can be protected under the Copyright Law.
Insufficient Evidence from the Plaintiff
The court held that the plaintiff’s evidence was clearly insufficient: 1. They failed to submit the original process records showing how the disputed image was generated in the AI software.2. The provided “describe” result was merely a post hoc description and could not reproduce the prompts used during generation.3. The so-called “reproduction description” was only a simulated attempt afterward, and the software/hardware environment, operation commands, and steps differed from those at the original time, so it could not prove that Zhou invested original intellectual labor during the initial generation.
Meanwhile, the reproduced result differed in style and composition from the disputed image, making it difficult to prove identity of the generation process. Overall, the current evidence was insufficient to determine that the disputed image had originality, and therefore it did not constitute a “work” under the Copyright Law.
The court of first instance dismissed all of Zhou’s claims. Zhou appealed, but the court of second instance upheld the original judgment, which has now taken legal effect.
The judge pointed out that although the number of intellectual property cases involving AI and big data is currently small, they are growing rapidly. For determining the originality of AI-generated works, the general principle of “he who asserts must prove” should be adhered to. Since AI generation requires lower intellectual input from natural persons compared to traditional painting, determining its originality must depend on case-specific evidence. Creators can protect their rights in the following ways: 1. When using AI to generate content, creators should retain complete generation records such as prompts, iteration processes, selection and modification records, and drafts to prove their intellectual labor and personalized expression.2. In addition to prompts, they may also preserve version comparisons, generation logs, screen recordings, and software parameter settings as objective evidence for use in ownership disputes.
In addition, industry actors should enhance the traceability capabilities of AI models and cooperate with institutional frameworks such as the Administrative Measures on Identifying AI-Generated and Synthesized Content, to prevent abuse of the copyright system and protect genuine innovation.
Comment
This case is one of the first domestic precedents to clearly establish standards for determining copyright in AI-generated images. Its judicial reasoning serves as a reference for future intellectual property protection in the AIGC field. It reminds creators that recognition of copyright does not depend solely on the finished product, but also on whether the “intellectual input” in the creative process can be proven. In the context of rapid AIGC development, building an evidence-conscious mindset of “leaving traces of the process” will be a key prerequisite for claiming copyright.