Mengchun’s Comments
This case raises a practical question: how should a court decide whether a disputed image was actually generated by AI?
The Guangzhou Internet Court accepted that AI detection reports can be submitted as evidence, but it did not treat the reported percentage as conclusive. The reason is straightforward: most detection platforms do not fully disclose their standards, algorithms, training data, or error rates. A report may therefore suggest that an image was AI-generated, but it cannot prove that fact on its own. It needs to be supported by other evidence.
Once the available evidence shows that an image was highly likely to have been generated by AI, the copyright claimant must do more to prove originality.
The court also made clear that using AI tools does not automatically rule out copyright protection. What matters is whether the final image reflects enough human creative input and individual choice. For that reason, creators of AI-assisted works should keep records of the creative process, such as PSD or other editable source files, layer changes, prompts, generation records, version histories, and later human edits. A finished image, a publication screenshot, or a simple screenshot of image layers may not be enough.
Chinese courts have also considered the separate question of whether AI-generated content can qualify for copyright protection. In a March 19, 2025 ruling, the Zhangjiagang Court held that content generated through simple prompts, without sufficient human intellectual input, did not qualify as a copyrightable work. For our earlier discussion of this issue, see: Jiangsu Zhangjiagang Court: AI-Generated Content Created Using Simple Prompts Does Not Constitute a Copyrightable Work (March 19, 2025 Ruling) – China IP Case Updates.
Case Background
The plaintiff claimed to have acquired the copyright in a Dragon Boat Festival-themed illustration from its original creator through an assignment agreement. The plaintiff alleged that a university infringed its right of communication through information networks by using the illustration in an article published on its official WeChat account without authorization.
The defendant argued that the illustration was highly likely to have been generated by artificial intelligence. According to multiple AI detection reports, the image showed an AI-generation probability of 99.65%. The defendant further submitted AI-generated images created using mainstream AI tools with similar prompts, asserting that the disputed illustration shared similar artistic style, composition, character design, and even identical visual anomalies, including abnormal or missing fingers.
During the proceedings, the court added the alleged original creator as a third party. However, the creator failed to appear before the court and did not provide any explanation or evidence regarding the creation process of the illustration.
Court’s Findings
The Copyright Owner Failed to Prove the Creative Process
The plaintiff submitted a copyright registration certificate, a screenshot showing the first publication of the illustration, the assignment agreement, payment records, and screenshots of layered image files.
The court held that these materials demonstrated the publication, registration, and transfer of the illustration, but did not sufficiently establish how the image had actually been created, who specifically created it, or whether the creation involved individualized human choices.
The court further explained that, for digital artwork, original editable source files—such as PSD files, layer files, operation logs, and version histories—are generally more capable of demonstrating the creative process than published image files. In this case, although the plaintiff submitted screenshots showing multiple layers, the screenshots did not contain the original source file information, layer modification history, creation time, image parameters, or other information capable of demonstrating drafting, line drawing, colouring, or subsequent editing. After the court requested additional evidence, the plaintiff stated that it was unable to contact the original creator and therefore could not provide further evidence regarding the creation process.
AI Detection Reports May Be Admitted but Have Limited Probative Value
The defendant submitted AI detection reports together with trusted timestamp evidence documenting the testing process. Multiple tests consistently indicated an AI-generation probability of 99.65%.
The court held that AI detection reports may be admitted into evidence. However, they are not equivalent to formal expert evidence obtained through court-recognized procedures. They are better treated as technical opinions submitted by one party and examined as private documentary evidence.
When assessing their admissibility, courts should consider factors including whether the tested image is clearly identified, whether the source of the tested material is reliable, whether the testing process has been properly preserved, whether the selected detection platform is relatively neutral, and whether any relationship exists between the platform and the party submitting the report.
As to their evidentiary weight, the court emphasized that current AI detection platforms generally do not fully disclose their detection standards, qualifications, algorithmic models, training data, or error-control mechanisms. Accordingly, AI detection reports may indicate the possibility that an image was AI-generated but cannot, standing alone, establish the origin of the image.
AI Detection Reports Must Be Corroborated by Other Evidence
In addition to the AI detection reports, the defendant submitted AI-generated images produced by several mainstream AI platforms using similar prompts, together with similar images obtained from social media.
The court found that these images shared substantial similarities with the disputed illustration in artistic style, character appearance, composition, and background settings. Some AI-generated images also displayed the same obvious physiological errors, including abnormal or missing fingers.
The court further observed that the disputed illustration itself contained similar anatomical anomalies. In addition, the social media account associated with the publication of the disputed illustration indicated that the publisher possessed AI image-generation capabilities and identified AI-generated content as one of the account’s characteristics.
Taken together, these materials supported the conclusion that the disputed illustration was highly likely to have been generated by AI.
The Burden Shifted Back to the Copyright Claimant
The court held that the plaintiff bore the burden of proving that the disputed illustration satisfied the requirements for copyright protection.
Once the defendant produced AI detection reports together with corroborating evidence indicating a high probability of AI generation, the plaintiff was required to provide further evidence of the creative process, the extent of human participation, individualized creative choices, and the formation of the final expression.
Neither the plaintiff nor the alleged original creator produced such evidence.
Accordingly, the court held that the plaintiff failed to establish that the disputed illustration constituted a copyrightable artistic work under the Copyright Law and dismissed all claims.
Judgment
The Guangzhou Internet Court dismissed all of the plaintiff’s claims. The judgment has become legally effective.
Case Reference
Court: Guangzhou Internet Court
Decision Date: 2026
Source: 【法脉准绳】AI率99.65%，端午插画还能主张版权吗？
