ABSTRACT

1 INTRODUCTION

Artificial intelligence (AI) is no longer an unfamiliar concept in contemporary scientific and technological development. The term “AI” (Artificial Intelligence) was first introduced by John McCarthy in 1956 at a conference held at Dartmouth College (New Hampshire, United States).¹Over nearly seventy years, AI has evolved to produce outputs of a high level of complexity that can be compared to human creative capacity. In Artificial Intelligence: A Modern Approach, the authors identify four core characteristics of modern AI: (i) thinking humanly, (ii) thinking rationally, (iii) acting humanly, and (iv) acting rationally.²Some scholars have even suggested that AI in the future may become more intelligent and powerful than the most intelligent human beings.³

At the end of 2024, the artwork AI God: Portrait of Alan Turing, painted by an AI robot, was sold at auction for over one million USD, setting a world record⁴and leading experts to recognise that AI-generated works may possess significant substantive and artistic value. Since 2021, the emergence of creative tools such as Stable Diffusion, Midjourney, DALL·E, and Dreamwriter has enabled users to generate images, music, and videos simply by entering natural-language prompts. This development has generated divergent views in legal scholarship as to whether AI-generated works should be recognised and protected by copyright in the same manner as works created by humans, and whether AI itself should be regarded as an author.

To address these legal issues, it is necessary to re-examine the understanding of works and copyright as regulated by national laws and relevant international treaties. In principle, copyright protection arises automatically without the need for any administrative formalities.⁵A work is eligible for copyright protection when it satisfies the following conditions. First, the work must be expressed in a particular form. The categories of protected works include all works in the fields of literature, science, and art, as provided for in Article 2 of the Berne Convention. Second, the work must be original. Originality is understood as independent creation, reflecting the author’s intellectual labour, and not copied from another work. Originality or creativity relates to the form of expression rather than merely the underlying idea of the work. Creativity does not require a work to be unique or to possess high artistic or academic value, but rather that it bears the personal imprint of the author, reflecting individuality, thought, emotion, and creative style.⁶Third, the work must not fall within the category of subject matter excluded from copyright protection. Each country has the authority to determine, under its domestic law, which subject matter is excluded from copyright protection, such as legal normative documents or news of a purely factual nature.⁷

The author who creates a work must be a natural person. This principle has been affirmed in international treaties, as well as in the WIPO’s guidance,⁸and in the practice of certain member states when encouraging copyright registration. For example, in notices concerning copyright ownership, the United States Copyright Office (USCO) emphasises that only a human being may be recognised as an author, and that copyright owners may be natural persons, legal persons, or unincorporated organisations,⁹without listing AI as a recognised author.¹⁰This approach is consistent with the understanding of authorship under the Chinese Copyright Law and reflects the general trend among states not to recognise AI as a subject of copyright.

However, the refusal to recognise AI as an author does not necessarily imply the denial of protection for AI-generated works. Whether copyright protection should be granted to works generated by AI remains an unresolved issue and continues to provoke considerable debate in legal scholarship. In this context, judicial interpretation and adjudicative practice may address legal gaps in the protection of copyright in AI-generated works.

2 METHODOLOGY

The study adopts a doctrinal legal research method to examine existing legal provisions and relevant legal practice concerning the protection of copyright in AI-generated works. It analyses international treaties and WIPO guidance, as well as the laws of selected jurisdictions, including the United States, the European Union, the United Kingdom, and China, to identify similarities and differences in national approaches to AI-generated works and the potential development of relevant legal frameworks. The United States, the European Union, and the United Kingdom are selected for in-depth analysis due to their leading roles in global investment in and application of artificial intelligence. The legislative positions and regulatory approaches adopted in these jurisdictions are likely to shape future trends in the development of AI-related legal frameworks.

In addition, the study conducts an in-depth analysis of notable judgments issued by the Chinese Internet Courts. These courts are specialised judicial bodies designed to adjudicate disputes arising in the digital environment through fully online proceedings. Notably, since 2023, cases involving copyright disputes over AI-generated works adjudicated by Chinese Internet Courts have attracted significant controversy both domestically and internationally. This development has opened up a novel analytical perspective in the study of legal protection for AI-generated works. In light of the existing legal gaps in copyright protection relating to AI, these courts have played a pioneering role in interpreting and applying the law to resolve disputes involving AI-generated works. By emphasising the role and perspectives of the Chinese judiciary, the article examines individual cases to evaluate judicial reasoning and to derive lessons for improving copyright law in the AI era.

The study is structured around the fundamental question of whether AI-generated works should be recognised as copyrightable. To address this issue, the article sequentially analyses and evaluates the legal positions adopted by selected jurisdictions in order to clarify prevailing approaches. Where legal gaps persist, judicial practice in China demonstrates how courts may contribute to the development of rules governing the protection of AI-generated works. Finally, based on the challenges identified and lessons drawn from judicial practice, the article proposes recommendations for establishing a comprehensive legal framework to protect copyright in AI-generated works, ensuring a balance between technological development and copyright protection.

3 LEGAL GAPS IN THE COPYRIGHT PROTECTION OF AI-GENERATED WORKS

3.1. The Distinction Between AI-Generated Works and AI-Assisted Works

Works created with AI involvement may be broadly divided into AI-generated and AI-assisted works. AI-assisted works refer to works in which natural persons use AI merely as a supporting tool in the creative process, such as refining images, suggesting language, editing texts, or assisting with music arrangement. This model typically follows the “prompt – input – output” workflow. In such cases, the core creative contribution and responsibility for the content remain with humans; therefore, these works may still qualify for copyright protection.¹¹The principal legal issue lies in determining the scope of human creativity relative to the extent of AI assistance, and whether the human contribution is sufficient to meet the originality requirement under copyright law. This issue poses significant challenges for legal systems when deciding whether AI-assisted works should be recognised as copyrightable works.

In contrast, AI-generated works are works produced entirely by AI systems without direct human intervention during the creative process. For example, an image generated by an AI model based solely on input commands such as “paint a row of cherry blossom trees in the spring breeze, with lightly drifting petals, soft colours, and a classical style,” without any subsequent human modification. In such cases, although AI technically synthesises data and generates a complete artistic work, it lacks legal authorship status under the copyright laws of most countries.¹²

However, where AI-generated works result from long-term human training of AI systems, through continuous data input and deliberate shaping of the information processed by AI to achieve a desired creative outcome, the question arises as to whether the trainer may be regarded as the author, and whether the AI-generated output may qualify as a “work” in the legal sense. In jurisdictions that do not recognise copyright protection for AI-generated works, it may be argued that “training AI” constitutes a technical or organisational activity rather than a direct creative act.¹³Consequently, the AI trainer would not be recognised as the author, and the AI-generated work would not be protected by copyright. Conversely, more open approaches argue that long-term AI training involves substantial creative investment, including data selection, parameter configuration, input adjustment, and training algorithm design, thereby reflecting personal intellectual input and creativity. Under this view, it may be appropriate to consider recognising the AI trainer as the author and granting copyright protection to the AI-generated work.

While the widespread use of AI offers undeniable benefits, it also poses significant legal challenges for copyright protection, particularly for AI-generated works. In practice, countries adopt divergent approaches. Under the expansive approach, in which a human selects data, designs algorithms, configures parameters, and intervenes to guide AI outputs toward a specific goal, such involvement may constitute creative activity bearing personal characteristics and thus warrant copyright protection. China, the United Kingdom, Singapore, and Japan are prominent examples of jurisdictions adopting this approach.¹⁴Under the traditional approach, a work must result from human intellectual creative labour; works generated entirely by AI do not satisfy copyright requirements unless humans input their own copyrighted works and can demonstrate effective control over the output.¹⁵The United States, the European Union, and Vietnam currently follow this approach.

No country currently prohibits AI from generating works; however, granting copyright protection AI-generated works remains a complex legal issue faced worldwide. For example, the USCO has stated that copyrightable works must be created by a human being, protecting only intellectual labour founded in the creative powers of the human mind.¹⁶In Australia, the case Acohs Pty Ltd v Ucorp Pty Ltd held that works created with computer intervention are not protected by copyright because AI is not a human author.¹⁷

In terms of form, early AI development focused on robots and tangible machines designed to perform repetitive mechanical tasks.¹⁸In recent years, AI has been embedded in software systems with increasingly sophisticated functions, such as facial recognition, data analysis, and predictive analytics with remarkable accuracy. Regarding diversity, AI has penetrated literature, arts, and other creative fields traditionally regarded as uniquely human domains. Even in complex areas such as legal research, AI can synthesise information, eliminate redundant data, improve legal file management, and effectively support legal consulting and litigation.¹⁹Based on machine learning technologies, AI processes input knowledge,²⁰often existing works, information, and knowledge available online, and applies complex algorithms to generate new outputs. AI-generated works may include texts, images, or musical compositions, some of which possess significant cultural and commercial value.

3.2. Legal and Regulatory Gaps in the Copyright Protection of AI-Generated Works in Selected Jurisdictions

3.2.1. Legal and Judicial Practice in the United States

The USCO has repeatedly held that images generated by AI are not eligible for copyright protection because they are not created by humans. Notable cases include:

Monkey selfie case (Naruto v. Slater, 2016): The copyrighted photograph at issue in this dispute was not created by AI but was taken by a monkey.²¹Nevertheless, both the court and the USCO affirmed that the photograph was not protected by copyright because a monkey cannot be an author and only human beings may qualify as authors of a work.²²This case has been regarded as an important precedent in subsequent debates on AI, as the USCO rejects the recognition of any non-human entity as an author.

Stephen Thaler and “a recent entrance to paradise” (Thaler v. Perlmutter, 2019): Thaler used an AI system known as the “Creativity Machine” to generate an artwork. The USCO denied copyright protection due to the absence of human authorship. This decision was upheld after Thaler initiated legal proceedings seeking to overturn the refusal between 2020 and 2025.²³In this case, Thaler’s involvement was limited to providing prompts and specific instructions expressing his intent for the AI to produce the final work based on given inputs. The degree of human intervention and labour was therefore considered minimal, while the AI carried out almost the entire process of producing the final output.

Zarya of the Dawn (2022): This case involved a graphic novel whose images were generated using Midjourney.²⁴The USCO granted copyright protection to the script, layout, and creative choices made by the human author, Kristina Kashtanova, but refused protection for the AI-generated images produced on the basis of the author’s script.²⁵This decision reaffirmed the USCO’s position that only human authors are recognised and that copyright protection extends solely to works directly created by humans.

U.S. courts have consistently upheld the USCO’s refusals, concluding that works generated by AI under human guidance or training, where human creativity is not concretely identifiable, are not eligible for copyright protection. U.S. courts maintain that the recognition of AI authorship or copyrightability of AI outputs is a matter for Congress. Consequently, U.S. administrative and judicial decisions adopt a narrow view,²⁶whereby, AI trainers or prompt providers are not considered authors unless they can demonstrate precise and predictable control over all aspects of the final output.

The United States has not enacted specific legislation addressing copyright protection for AI-generated works, aside from policy reports issued by the USCO. The U.S. Copyright Act protects original works of authorship fixed in a tangible medium and created by humans, while the Digital Millennium Copyright Act (DMCA 1998) does not address AI-generated works.²⁷According to USCO guidance, copyright protection applies only if it is the result of intellectual labour, which can be achieved solely through the creative capacity of the human mind.²⁸As a result, applications for works not created by humans are uniformly rejected.

Overall, the United States continues to adhere to a traditional approach that requires clear, and direct human creativity. Even extensive AI training is regarded as a technical activity rather than creative labour, resulting in persistent legal gaps as AI increasingly produces works comparable in value to human-created literary, scientific, and artistic works.

3.2.2. Legal and Judicial Practice in the European Union

The European Union has demonstrated rapid adaptability to technological development through the adoption of the AI Act,²⁹approved by the European Parliament in June 2024. The AI Act aims to establish a regulatory framework for AI governance and mitigate risks arising from AI misuse by classifying AI systems according to their risk levels to human health and fundamental rights.³⁰

However, the AI Act primarily proposes that individuals or entities making significant contributions to AI-generated works may be entitled to copyright protection, without specifying concrete criteria or thresholds for significant contribution. Its primary role lies in ethical boundaries, risk awareness³¹and responsible AI development. Similar to the United States, the EU continues to uphold the principle of human intellectual creativity and lacks a clear legal mechanism to recognise AI trainers as authors. Nevertheless, the AI Act reflects the EU’s proactive legislative stance and opens for the door to future copyright protection for AI-generated works.

In judicial practice, EU courts do not recognise non-human authorship and remain reluctant to protect works created entirely by AI.³²In Infopaq International A/S v. Danske Dagblades Forening (2009), the Court of Justice of the European Union (CJEU) held that copyright applies only to original works reflecting the author’s own intellectual creation, a standard that AI-generated works cannot satisfy.³³CJEU case law has played a crucial role in harmonising the interpretation of intellectual property across EU member states,³⁴thereby contributing to a coherent copyright framework in the context of rapid technological advancement.

EU intellectual property offices have rejected applications listing AI as the author³⁵while leaving open the possibility of protection where a human author is identified and the AI-generated work demonstrates originality. Although AI authorship is not recognised, the EU does not entirely preclude future copyright protection for AI-generated works.

3.2.3. Legal and Judicial Practice in the United Kingdom

The United Kingdom adopted an early approach to AI-related intellectual property issues. Section 9(3) of the Copyright, Designs and Patents Act 1988 provides that³⁶for computer-generated literary, dramatic, musical, or artistic works, the author is the person by whom the arrangements necessary for the creation of the work are undertaken. Traditionally, this has led to identifying programmers or engineers as authors. In the contemporary AI context, this may extend to AI trainers,³⁷provided they demonstrate creative control.

In Nova Productions Ltd v Mazooma Games Ltd, the court held that images generated during gameplay constituted computer-generated works and that the software programmer was the author, as the programmer predetermined the rules and commands governing the creation of those images.³⁸Although players could trigger the appearance of images, their actions remained subject to the constraints embedded in the software.³⁹Similarly, in THJ Systems Ltd v Sheridan, where the court recognised copyright protection for images generated by software and identified the programmer as the author based on creative input and personal intellectual contribution, while the software developer was considered the copyright owner.⁴⁰However, judicial interpretation of Section 9(3) remains limited, and uncertainty persists. In 2024, the UK Intellectual Property Office indicated the possibility of repealing Section 9(3),⁴¹though no definitive decision has been made due to the absence of a comprehensive legal framework governing AI-generated works.

Section 178 of the same Act defines computer-generated works as works with no human author, suggesting possible copyright protection.⁴²Nonetheless, concerns arise over AI’s reliance on existing copyrighted works as training data, risking infringement and undermining originality.⁴³Expanding copyright protection may conflict with the fundamental principles of intellectual property, as protected works risk being exploited as mere inputs for further AI-generated outputs. Consequently, ethical and legal challenges surrounding the recognition of copyright in AI-generated works remain unresolved, raising unresolved ethical and legal risks in the UK.

4 JUDICIAL PRACTICE OF CHINA’S INTERNET COURTS IN COPYRIGHT DISPUTES INVOLVING AI-GENERATED WORKS

Chinese law does not recognise AI or non-human entities as authors. Whether AI-generated works qualify for copyright protection remains unclear under the law. Nevertheless, China has adopted an open yet cautious approach through the judicial practice of its Internet Courts. The first Chinese Internet Court was established in Hangzhou in 2017, and similar courts were established in Beijing and Guangzhou in 2018.⁴⁴The jurisdiction of Chinese Internet Courts is to resolve disputes related to data ownership, privacy protection, violations of virtual property rights, and unfair competition in cyberspace.⁴⁵Cases resolved by China's internet courts may offer an initial glimpse into legal thinking regarding artificial intelligence⁴⁶and represent a pioneering step by China toward creating a judicial system capable of handling issues arising from the digital economy.

4.1. “Spring Breeze Brings Tenderness” Case (Li v. Liu, 2023)

In February 2023, the claimant used Stable Diffusion to generate a portrait titled Spring Breeze Brings Tenderness⁴⁷and posted it on Xiaohongshu (Rednote).⁴⁸The defendant reproduced the image on a Baidu-affiliated platform without authorisation, removed the watermark and used it without permission. The claimant brought an action before the Beijing Internet Court, alleging copyright infringement.

The court addressed two key issues: (i) whether the AI-generated image constituted a work under copyright law, and if so, what type of work; and (ii) whether the claimant held copyright and whether the defendant’s conduct amounted to infringement.

On the first issue, the court held that the claimant had engaged in intellectual activity throughout the creative process, including conceptualisation, selection of training inputs, repeated adjustment of prompts and parameters, and final selection of the image.⁴⁹Accordingly, the court classified the image as an artistic work. On the second issue, the court ruled that the AI model developer merely provided a creative tool and could not be regarded as the author. Given that the claimant independently used Stable Diffusion and selected the final image, the court recognised the claimant as the author and copyright holder. In its judgment on 27 November 2023, the Beijing Internet Court found that the defendant had infringed the claimant’s copyright, ordered a public apology, and awarded economic compensation.

This case marked the first judicial decision in China to affirm copyright protection for an AI-generated image and was listed among the top ten nominated cases of 2024.⁵⁰It reflects a creativity-oriented approach consistent with China's copyright law, which aims to encourage creation and dissemination of works.⁵¹Although the statute does not explicitly address AI-generated works, the court interpreted its underlying principles in light of technological development.

However, the decision attracted substantial domestically and international criticism. Opponents argued that, despite human involvement in controlling inputs, the final output was autonomously generated by AI through processes beyond ordinary human control.⁵²Equating AI image generation with human creativity was therefore considered conceptually flawed,⁵³and AI outputs were viewed as technical products rather than copyrightable works.

4.2. “Cat Crystal Diamond Pendant” Case (2025)

The “Cat Crystal Diamond Pendant” case concerns a copyright infringement dispute before the Beijing Internet Court. The claimant, Zhou, a content creator, alleged that he independently created an AI-generated image titled “Cat Crystal Diamond Pendant” using image-generation software (presumably Midjourney). He claimed that, during a business collaboration, the defendant, a Beijing technology company, disseminated the image in a WeChat group without authorisation.

On 16 September 2025, the Beijing Internet Court issued its decision in this case. According to this decision, the court rejected the claimant’s claim, holding that the image did not qualify as a copyright-protected work.⁵⁴

The court acknowledged that AI-generated images may, in principle, be protected, but only where the claimant can demonstrate substantial creative input and personal expression. Claimants must explain their creative intent, the prompts used, the selection and adjustment of inputs, and provide supporting evidence. In this case, the court found that the claimant’s contribution was limited to providing general ideas and prompts, while the remaining process was completed by AI with minimal intellectual effort.⁵⁵Compared to traditional manual creation, the level of human creativity was insufficient, and copyright protection was denied.

Taken together, these two cases illustrate the cautious stance of Chinese internet courts. Rather than granting blanket protection to AI-generated outputs, courts assess the degree of human creative involvement,⁵⁶thereby preserving the core objective of copyright law: promoting genuine creativity and the dissemination of works beneficial to social development.

5 SHIFTS IN THE PERCEPTION OF COPYRIGHT IN AI-GENERATED WORKS

5.1. Originality and ethical concerns in art

The primary reason many jurisdictions refuse to recognise copyright in AI-generated works lies in disputes over originality. Authors are generally required to demonstrate intellectual labour, creative judgment, and personal imprint. In the case of AI-generated works, it is difficult to clearly distinguish between pre-existing scientific knowledge, protected works used as training data,⁵⁷and individual creativity. As a result, AI outputs are often perceived as copies or compilations rather than original works bearing distinct personal expression. Moreover, AI technologies frequently exploit existing creative works as training data without authorisation,⁵⁸raising ethical and legal concerns. The large-scale aggregation and replication of content by AI models also increases the risk of replacing human artists in creative industries.⁵⁹

At the same time, the rapid development of AI suggests that such limitations may be partially overcome. AI systems may incorporate original ideas introduced by developers or trainers. Based on its ability to self-learn and research, AI may generate novel algorithms and outputs not previously known to humans.⁶⁰In addition, human intervention at the refinement and publication stages may embed sufficient personal input to satisfy copyright requirements. Given the significant commercial value generated by AI-assisted works,⁶¹recognising their economic contribution could facilitate reinvestment and expand creative opportunities.

At present, WIPO has not adopted binding rules on copyright protection for AI-generated works, but it has actively facilitated international dialogue on this issue.⁶²Discussions focus on whether AI outputs meet copyright standards, who should be recognised as author or rights holder, and how AI-driven creativity may affect traditional artistic norms. Although no consensus has emerged, WIPO continues to gather views from member states and experts before proposing any regulatory framework. Furthermore, WIPO could take the lead in proposing amendments to the Bern Convention, reconsidering the approach and interpretation of the concept of an original work in the age of technology, thereby moving towards accepting AI-generated works within the Bern Convention's scope.

5.2. The Subject of Copyright

The global legal framework governing AI and AI-related intellectual property remains at an early stage. Most jurisdictions neither recognise copyright in AI-generated works nor accept AI as an author or rights holder. This position reflects uneven technological development and divergent national approaches to AI governance. Fundamentally, AI is regarded as technology rather than a natural person,⁶³and therefore lacks the capacity to bear legal rights and obligations.

Although AI systems may operate autonomously, they remain dependent on human-created algorithms and knowledge. Consequently, AI is generally treated as a tool rather than an independent subject of law, and does not satisfy the requirement of direct creative authorship.⁶⁴The prevailing international trend is therefore to allocate copyright in AI-generated works to human or legal persons involved in developing, owning, or operating AI systems.⁶⁵This approach aligns with existing legal structures and the objective of protecting human intellectual achievements, while also incentivising investment and innovation.

Nevertheless, assigning copyright to human or corporate actors may create risks of abuse, such as excessive licensing fees or overreliance on advanced technology, potentially undermining the public interest. Balancing economic incentives and access remains a key reason why neither states nor WIPO have reached a definitive position on copyright protection for AI-generated works.

5.3. Policy Recommendations

The AI era presents unprecedented legal challenges, requiring a departure from traditional doctrines. China’s evolving judicial practice reflects both caution and openness in addressing issues not yet clearly regulated by law. Although current court decisions function as precedents rather than statutory rules, they demonstrate flexibility in interpreting creativity and recognising human intellectual contribution through AI.

Drawing on the judicial practice analysed above, China’s approach clearly encourages technological innovation. Rather than rigidly applying existing copyright rules to AI-generated works, Chinese courts have gradually expanded the concept of creativity by recognising human intellectual contribution through the training, supervision, and control of AI systems capable of producing copyrightable works. Assessments of creative judgment, intellectual labour, originality, and personal imprint remain highly dependent on judicial discretion in individual cases.⁶⁶While this flexibility allows courts to respond to technological realities, it also risks inconsistency in the absence of a comprehensive statutory framework. However, case-by-case assessment of originality and creative input, largely dependent on judicial discretion, risks inconsistency in the absence of clear legislation. To address these challenges, coordinated action at national and international levels is necessary.

First, legislation should clarify the legal status of AI. As a scientific and technological phenomenon, AI must be legally characterised as a specific entity, whether tangible or intangible. At present, AI may be conceptualised either as: (i) a subject of law, with limited recognition of rights in rare cases,⁶⁷such as Saudi Arabia’s approach to humanoid robots;⁶⁸or (ii) a distinct legal object subject to special regulatory control. Under the latter approach, AI is not recognised as a rights-bearing entity but rather as a controlled asset or technological tool,⁶⁹as reflected in Article 3(1) of the EU AI Act. The EU regulations approach AI as a distinct legal object, controlled by subjects of law and subject to separate control, to limit risks to AI users. Clearly defining the legal nature of AI is crucial for assessing originality and creativity in AI-generated works, and directly affects decisions on copyright eligibility.

Second, international treaties and national laws could reconsider the concept of authorship by shifting from a narrow definition, in which the author is the person who directly creates the work, to a broader understanding of authorship as a creative contribution. This would enable recognition of copyright for individuals who develop, train, or meaningfully use AI systems, provided they can demonstrate substantial intellectual investment and personal creative input. At the same time, ownership rules may be expanded to include investors such as AI system owners, data holders, or publishers who finance AI-assisted production. This approach would enhance economic incentives, stimulate innovation,⁷⁰and support the sustainable development of the AI industry.

Third, specific criteria should be established to assess the copyright protection of AI-generated works. Unlike traditional works, AI-generated outputs require tailored standards. Originality should remain central, requiring evidence of intellectual effort and personal imprint.⁷¹AI-generated works should not merely reorganise existing data but must demonstrate novelty supported by identifiable creative processes. Rights holders should be able to document initial intellectual investment, creative planning, and the continuous involvement of human actors. Currently, there are no specific standards for quantifying the author's contribution in proving the above points; it depends on each case handled by the court. As reflected in Chinese internet court decisions, such criteria should be codified in legislation rather than applied only in isolated cases, ensuring consistency and objectivity in future disputes. Therefore, in line with judicial practice, courts may submit opinions to the legislature to develop a set of evaluation standards or a classification system for determining the originality of AI-generated works.

Fourth, AI training relies heavily on large volumes of copyrighted works owned by third parties. Unlike traditional quotation practices, AI developers may extensively reuse protected content without permission or remuneration.⁷²Domestic intellectual property laws should, therefore, require AI enterprises to provide fair compensation to copyright holders for the use of training data. This mechanism would safeguard existing rights while enabling AI development within a transparent and lawful framework. One feasible approach would be to operationalise compensation through a collective licensing model administered by Collective Management Organisations (CMOs). Although CMOs have expressed concerns that AI enterprises are currently evading payment obligations to copyright holders,⁷³increasing cooperation among CMOs across jurisdictions indicates emerging prospects for the development and institutionalisation of such licensing models in the AI era.

6 CONCLUSIONS

AI and the legal challenges it raises for AI-generated works have become a major concern worldwide. As such, works that continue to improve in both substance and form significantly reshape perceptions of science and technology. While embracing the benefits of AI development, states must also confront its negative implications. In the field of copyright, AI enables the rapid creation of high-quality works, yet raises a fundamental question: should AI-generated works be treated on par with human-created works in literature, science, and the arts? Even technologically advanced jurisdictions such as the EU and the United States remain cautious in defining AI’s legal status and the originality of AI-generated works. As a result, copyright applications for AI-generated works are often rejected, and domestic laws have yet to fully recognise them as equivalent to works created by humans.

Judicial practice of the Chinese Internet Courts demonstrates that copyright protection for AI-generated works largely depends on whether the claimant can prove meaningful creative effort and a discernible personal imprint in the work. Given AI’s capacity for self-learning and algorithmic optimisation, its outputs may be substantially shaped, refined, and developed through human intervention prior to publication. Such an intervention can embed sufficient human creativity to satisfy the requirements of copyright protection.

Ultimately, copyright protection for AI-generated works in the technological era is not merely a legal issue but also a strategic policy choice. States must strike a balance between protecting intellectual property rights, encouraging innovation, and managing technological risks to prevent the misuse of AI. In the future, addressing legal uncertainties and establishing a coherent framework for protecting AI-generated works will be essential to safeguarding the interests of creators and technology investors while promoting sustainable economic development in the digital age.

FOOTNOTES

REFERENCES

AUTHOR’S INFORMATION

RIGHTS AND PERMISSIONS

EDITORS

ABOUT THIS ARTICLE

DETAILS FOR PUBLICATION

AI DISCLOSURE STATEMENT

ACKNOWLEDGEMENT

АНОТАЦІЯ

中文注释*⁷⁴