1. Introduction

Artificial Intelligence (AI) has fundamentally changed the way intellectual property (IP) is created, owned, licensed, and protected. AI systems can generate literary works, paintings, music, software code, inventions, product designs, and other outputs that traditionally resulted from human creativity or inventive activity.

These developments raise several important legal questions:

Can an AI system be recognised as an author or inventor?

Who owns copyright in content generated by AI?

Can an AI system or its developer obtain patent rights?

Does training an AI model on copyrighted material constitute infringement?

Who is liable when AI-generated material copies protected works?

How should legal systems balance innovation, human creativity, commercial interests, and public access to knowledge?

Civil law and comparative AI intellectual property governance examines how different legal traditions and jurisdictions respond to these questions through legislation, judicial decisions, administrative regulations, and international principles.

A central distinction is that civil law can refer both to a legal tradition based primarily on codified statutes and to private law governing relationships between individuals and organisations. In this discussion, the term primarily refers to the civil-law legal tradition, particularly in continental Europe, while also considering common-law jurisdictions such as the United States, United Kingdom, and India.

The central challenge is to adapt established intellectual property principles to a technology in which creative and inventive processes may involve varying degrees of human control and machine autonomy.

2. Meaning and scope of AI intellectual property governance

AI intellectual property governance refers to the legal and institutional framework regulating the creation, ownership, use, protection, licensing, and enforcement of intellectual property involving AI systems.

It covers two distinct but interconnected areas.

A. AI as a creator or inventor

Whether AI-generated works and inventions qualify for copyright, patents, industrial design protection, or other IP rights, and who can own those rights.

B. AI as a user of protected material

Whether the collection, copying, ingestion, training, storage, and output of AI systems infringe the rights of authors, artists, publishers, software developers, or patent holders.

These questions must be analysed separately. A work may be generated by an AI system without qualifying for copyright protection, while the training process used to create that system may independently raise copyright questions.

Similarly, an AI-assisted invention may be patentable if it satisfies the relevant legal requirements, even though the AI system itself cannot be named as the inventor.

3. Foundational legal principles

3.1 Human authorship

Traditional copyright law generally assumes that a work originates in human intellectual activity. The precise legal test varies by jurisdiction, but human creativity, intellectual contribution, or original expression commonly plays a central role.

Where a person uses AI as a tool and exercises meaningful creative control over the resulting work, copyright may protect the human-authored elements. Where the output is generated autonomously with no legally sufficient human creative contribution, protection may be unavailable in jurisdictions that require human authorship.

3.2 Originality

Originality generally requires that a work meet the applicable threshold of independent creation and legally recognised creative expression.

AI-generated material creates a difficult question: does the originality lie in the output itself, the human prompts, the selection and arrangement of outputs, or subsequent editing?

A prompt alone will not necessarily establish authorship. The relevant inquiry is whether the human's contribution satisfies the jurisdiction's substantive originality and authorship requirements.

3.3 Inventorship and patentability

Patent law generally distinguishes between:

Inventorship: identification of the person or persons who legally qualify as inventors.

Patentability: whether an invention satisfies requirements such as novelty, inventive step or non-obviousness, and industrial applicability or utility.

Ownership: who holds the resulting patent rights, which may be assigned to an employer, company, or another legal person under applicable law.

An AI system may contribute to the technical development of an invention, but that does not automatically make the system a legally recognised inventor.

3.4 Territoriality

Intellectual property rights are principally territorial. A work may receive protection in one jurisdiction but face different treatment in another because the applicable statutes, judicial interpretations, exceptions, and ownership rules differ.

Consequently, an AI-generated image distributed globally may have different copyright implications in the United States, India, the European Union, and the United Kingdom.

3.5 Technological neutrality and accountability

A sound governance framework should regulate legally relevant conduct rather than assume that every AI-generated output is either protected or infringing.

It should also identify the relevant actors: developers, model providers, deployers, users, data suppliers, and commercial distributors. Their legal responsibilities depend on the activity performed, their knowledge and control, and the applicable statutory framework.

4. Comparative AI intellectual property governance models

Different jurisdictions follow different approaches to authorship, training-data use, patent inventorship, and accountability. For comparative purposes, six important models are the European Union, United States, United Kingdom, India, China, and Japan.

Checking Favorites status

4.1 European Union — rights-based and regulatory model

The EU combines copyright law with a detailed framework for regulating AI. Its approach places importance on authors' rights, transparency, lawful access to training material, and the regulation of certain AI providers.

Key instruments include the EU Copyright Directive 2019/790, the EU AI Act (Regulation 2024/1689), and national copyright legislation interpreted consistently with EU law.

Copyright and training: Articles 3 and 4 of the Copyright Directive establish text-and-data-mining exceptions under specified conditions. The scientific-research exception has particular safeguards, while the broader exception is subject to lawful-access requirements and, where applicable, effective reservations of rights by rightsholders.

AI governance: The AI Act introduces obligations for general-purpose AI model providers, including a copyright-compliance policy and a sufficiently detailed public summary of training content. These requirements do not themselves establish that every training use is lawful or confer copyright on AI-generated outputs.

Distinctive feature: A combination of copyright protection, regulatory oversight, and transparency obligations.

Checking Favorites status

4.2 United States — human-authorship and fair-use model

US copyright law generally requires human authorship. The US Copyright Office's 2025 report on generative AI outputs maintains that copyright protection depends on sufficient human-determined expressive elements; merely entering prompts is not, by itself, enough.

The principal statutory framework is the Copyright Act of 1976, particularly the originality, ownership, exclusive-rights, and fair-use provisions. Patent law separately identifies inventors under the Patent Act.

Training and infringement: Whether using copyrighted works to train an AI system constitutes fair use depends on the facts and applicable law. The analysis may consider the purpose of use, the nature of the works, the amount used, and the effect on the relevant market.

Distinctive feature: Human authorship for copyright, combined with case-by-case judicial analysis of infringement and fair use.

Checking Favorites status

4.3 United Kingdom — statutory computer-generated works model

The UK Copyright, Designs and Patents Act 1988 contains a specific provision, section 9(3), for computer-generated literary, dramatic, musical, or artistic works. Where a work is computer-generated in circumstances in which there is no human author, the author is deemed to be the person who undertook the arrangements necessary for its creation.

This provision differs from the more explicitly human-authorship-centred US approach. However, it does not automatically settle every question about generative AI, the identity of the person making the necessary arrangements, or the originality of the resulting work.

UK law also provides copyright exceptions relevant to computational analysis, including a limited text-and-data-mining exception for non-commercial research under section 29A.

Distinctive feature: A statutory rule addressing computer-generated works, alongside conventional copyright infringement rules and relatively narrow text-and-data-mining provisions.

Checking Favorites status

4.4 India — conventional IP law applied to emerging AI issues

India's Copyright Act 1957, Patents Act 1970, Trade Marks Act 1999, and Designs Act 2000 provide the principal statutory foundations.

Section 2(d)(vi) of the Copyright Act expressly addresses the author of a computer-generated literary, dramatic, musical, or artistic work by referring to the person who causes the work to be created. This provision is relevant to AI-generated content, although its precise application to modern generative systems remains legally unsettled.

Patent law presents a separate question: whether the legally recognised inventor must be a natural person and how an AI-assisted invention should be attributed. There is no general rule that gives an AI system independent patent ownership merely because it contributed to an invention.

Distinctive feature: Existing statutory provisions must be interpreted in the context of generative AI, with significant questions still requiring authoritative clarification.

Checking Favorites status

4.5 China — judicial recognition of human creative contribution

Chinese copyright law has been applied in disputes concerning AI-generated images. A significant example is the Beijing Internet Court's 2023 decision concerning an image generated using Stable Diffusion.

The court examined the user's selection of prompts, parameters, and iterative adjustments to determine whether the resulting image reflected sufficient intellectual input and originality.

China's broader governance environment also includes rules addressing generative AI services, algorithmic recommendations, and synthetic content. Copyright questions must be distinguished from these regulatory requirements.

Distinctive feature: Case-specific judicial assessment of human creative contribution, combined with a substantial regulatory framework for AI services.

Checking Favorites status

4.6 Japan — innovation-oriented text-and-data-mining model

Article 30-4 of Japan's Copyright Act generally permits certain uses of copyrighted works for purposes that do not aim to enjoy the thoughts or sentiments expressed in them, subject to statutory conditions and limitations.

This provision is important for machine learning because data analysis and model training may involve the use of protected works without the same purpose as ordinary consumption of those works.

However, the exception is not unlimited. Uses that unreasonably prejudice the interests of rightsholders may fall outside its protection, and the creation or distribution of infringing outputs raises separate issues.

Japanese law also distinguishes the use of AI as a tool from the question of whether an output embodies sufficient human creative expression to qualify for copyright.

Distinctive feature: A comparatively permissive statutory framework for certain analytical uses, while preserving limits to protect rightsholders.

Comparative overview

The table describes broad tendencies, not a complete statement of every national rule. In particular, AI training exceptions, ownership provisions, and procedural requirements vary within the EU and across national legal systems.

5. Case laws on AI and intellectual property governance

The following decisions illustrate how courts and administrative bodies have addressed authorship, inventorship, originality, training data, and infringement. They should be read in their particular procedural and jurisdictional context; not every decision determines the ultimate legality of AI training.

Case 1. Thaler v. Perlmutter — United States

Court: US Court of Appeals for the District of Columbia Circuit, 2025; affirming the district court.

Subject: Copyright in an autonomously generated image.

Facts: Stephen Thaler sought copyright registration for an image generated by his AI system, identifying the machine rather than a human being as the author. The US Copyright Office refused registration because the work lacked the required human authorship.

Legal issue: Can a work created autonomously by an AI system qualify for US copyright protection when no human is identified as its author?

Decision and principle: The courts upheld the human-authorship requirement. The statutory concept of an author, as interpreted by the court, did not encompass a machine as the author of the work in question.

Significance for AI governance:

AI cannot simply be named as a copyright author under the governing US rule.

Ownership of the machine or AI system does not automatically establish copyright in every output it generates.

Human contributions may still qualify for protection where they satisfy applicable requirements.

Critical analysis: The decision addresses the identity of the author, not every possible form of AI-assisted creation. It does not mean that all works involving AI are unprotected. A human-created composition, selection, arrangement, or modification may be protected even if some elements were generated by AI.

Case 2. Thaler v. Vidal — United States

Court: US Court of Appeals for the Federal Circuit, 2022.

Subject: AI inventorship under patent law.

Facts: Thaler attempted to list his AI system, DABUS, as the inventor of patent applications. The US Patent and Trademark Office rejected the applications on the ground that an inventor under the Patent Act must be an individual.

Legal issue: Can an AI system be named as an inventor in a US patent application?

Decision: The Federal Circuit held that an inventor under the relevant statutory language must be a natural person. The decision was subsequently left standing when the US Supreme Court declined to review the case.

Legal principles:

Inventorship is a legal status distinct from ownership of a machine.

A person or company may own a patent without being the inventor.

AI assistance does not automatically prevent patent protection, provided the legally required human inventorship and other patentability requirements are satisfied.

Comparative importance: This decision demonstrates that patent law and copyright law independently impose attribution requirements. The answer to whether AI-generated content can receive copyright protection does not itself determine whether an AI-assisted invention is patentable.

Case 3. Thaler v. Comptroller-General of Patents, Designs and Trade Marks — United Kingdom

Court: UK Supreme Court, 2023, 20232023 UKSC 49.

Subject: Patent applications naming an AI system as inventor.

Facts: Thaler filed patent applications identifying DABUS as the inventor. He argued that, as the owner of the machine, he was entitled to obtain the patents arising from its output.

Legal issue: Could DABUS qualify as an inventor under the Patents Act 1977, and could Thaler obtain patent rights through ownership of the machine?

Decision: The Supreme Court held that an inventor under the relevant legislation must be a natural person. Ownership of DABUS did not, by itself, establish a right to obtain a patent for an invention attributed to the machine.

Ratio and significance:

The statutory meaning of inventor could not be extended to an AI machine in this way.

Ownership of an AI system and entitlement to patent rights are distinct legal questions.

Patent applications must comply with statutory inventorship and entitlement requirements.

Comparative lesson: Although UK copyright law contains a specific provision for certain computer-generated works, UK patent law does not similarly treat an AI system as a legal inventor. Different IP regimes can therefore reach different results concerning the same technology.

Case 4. Commissioner of Patents v. Thaler — Australia

Court: High Court of Australia, 2022, 20222022 HCA 35.

Subject: Whether an AI system can be named as a patent inventor.

Facts: The dispute concerned patent applications naming DABUS as the inventor. The applicant argued that patent law should recognise an AI system as an inventor even though the system was not a human being.

Legal issue: Did the Australian Patents Act permit an AI system to be identified as the inventor?

Decision: The High Court rejected the appeal, leaving in place the conclusion that the statutory inventor must be a natural person under the applicable law.

Significance:

The court did not create a separate category of machine inventorship.

The decision reinforced the importance of statutory language in determining who may be an inventor.

The case illustrates that questions about whether AI can technically invent are distinct from whether AI can legally qualify as an inventor.

Comparative lesson: The Australian decision aligns with the US and UK approaches to inventorship, illustrating the influence of existing patent statutes even as technological capabilities evolve.

Case 5. Zarya of the Dawn — United States Copyright Office determination

Authority: US Copyright Office, registration determination and reconsideration, 2023.

Subject: Copyright in a comic book containing AI-generated images.

Facts: Kristina Kashtanova submitted a comic book incorporating text and images generated using Midjourney. The Copyright Office reconsidered the scope of protection after learning more about the process used to produce the images.

Legal issue: Can a human-authored work containing AI-generated images receive copyright protection, and which elements qualify?

Determination: The Office recognised protection for the human-authored text and the sufficiently creative selection, coordination, and arrangement of the work, while declining to recognise copyright in the AI-generated images themselves on the facts presented.

Legal principles:

Copyrightability must be assessed by examining the human contribution.

A larger work can qualify for protection even when some constituent material does not.

Registration does not necessarily extend to every visual or textual element included in a submission.

Significance: This determination provides a practical model for separating protected human expression from unprotected AI-generated material.

Practical implication: Creators should preserve records of their own writing, editing, compositional decisions, and other creative contributions, and accurately identify AI-generated elements when seeking registration.

Case 6. Beijing Internet Court AI-generated image decision — China

Court: Beijing Internet Court, November 2023.

Subject: Copyright in an image generated using Stable Diffusion.

Facts: A user generated an image using an AI image-generation system and published it online. A dispute arose when another person used the image without permission.

Legal issue: Could the AI-generated image be recognised as an original work protected by Chinese copyright law?

Decision: The court found copyright protection appropriate on the facts before it, taking into account the user's choices in prompts, parameters, and iterative adjustments that contributed to the resulting image.

Reasoning: The relevant issue was not simply that an AI system had produced an image. The court considered whether the user's intellectual effort and individual creative choices were reflected in the output to a sufficient degree.

Significance:

Human interaction with AI may, in suitable circumstances, contribute to a protectable work.

The amount and nature of the human contribution matter.

The judgment should not be read as declaring all AI-generated images copyrightable.

Comparative lesson: The decision contrasts with an approach that focuses primarily on whether the machine itself can be an author. It instead examines the user's creative input and the originality of the resulting work.

Case 7. Andersen v. Stability AI Ltd. — United States

Court: US District Court for the Northern District of California; significant pleadings rulings in 2023 and 2024, with proceedings continuing thereafter.

Subject: Alleged copyright infringement through training and distribution of generative image models.

Facts: A group of artists alleged that companies associated with Stable Diffusion and other image-generation systems had used their copyrighted works without authorisation in developing AI models and had generated outputs that implicated their rights.

Legal issues: The litigation raises questions about copying during training, the relationship between training datasets and model weights, direct and secondary infringement, and the sufficiency of allegations connecting particular works to particular defendants.

Procedural significance: The court's rulings allowed some claims to proceed while dismissing or narrowing others. Those preliminary decisions should not be mistaken for a final determination that AI training is always infringing or always fair use.

Importance for governance:

Training-data provenance may be central to establishing or contesting liability.

Copyright claims require legally sufficient allegations and evidence; the presence of a copyrighted work somewhere in a dataset does not automatically resolve every element of infringement.

Developers and rightsholders face difficult evidentiary questions concerning model training and output generation.

Critical analysis: The litigation illustrates why courts must distinguish the copying involved in acquiring and processing training material from any infringing reproduction, adaptation, or distribution of particular outputs. Those acts may raise separate legal questions.

Case 8. Getty Images v. Stability AI — United Kingdom

Court: High Court of England and Wales, Chancery Division, judgment of 4 November 2025, 20252025 EWHC 2863 (Ch).

Subject: Copyright, database rights, trade marks, and AI image-generation systems.

Facts: Getty Images alleged that Stable Diffusion had been developed using visual assets taken from websites, including material over which Getty claimed rights. It also challenged certain outputs and the use of its trade marks.

Legal issues: The case raised questions about the use of protected images in training, whether model weights constituted infringing copies, whether distribution of the model could attract secondary copyright liability, and whether generated images could infringe trade marks or amount to passing off.

Outcome and procedural context: By the trial, Getty had abandoned its principal training-and-development claim after acknowledging that there was no evidence that the relevant training and development occurred in the UK. The outputs claim was also substantially abandoned after the allegedly problematic prompts had been blocked. The High Court rejected the remaining secondary copyright infringement claim concerning the model's importation and distribution, on the evidence before it. Trade-mark and passing-off issues required separate analysis.

Significance:

Copyright infringement depends on the particular acts alleged and the relevant territorial law.

A model's parameters or weights are not automatically equivalent to copies of every work used in training.

Trade-mark infringement and passing off may arise independently of copyright.

The case illustrates the importance of evidence concerning where training occurred, what the model retains, and how it is deployed.

Comparative lesson: The judgment demonstrates the limits of treating AI training as a single legal act. Collection, copying, training, distribution, and output generation must be analysed separately.

Case 9. Naruto v. Slater — United States

Court: US Court of Appeals for the Ninth Circuit, 2018, 888 F.3d 418.

Subject: Non-human authorship and copyright ownership.

Facts: A crested macaque known as Naruto took photographs using a camera belonging to a wildlife photographer. An animal-rights organisation brought proceedings seeking to enforce copyright on the monkey's behalf.

Legal issue: Could a non-human animal claim copyright under US copyright legislation?

Decision: The Ninth Circuit held that the Copyright Act did not authorise the animal to bring the copyright claim in question.

Relevance to AI: The case is not about artificial intelligence, but it illustrates the difficulties of extending copyright authorship and enforcement rights beyond the legal categories recognised by legislation.

Critical distinction: An animal and an AI system are not legally identical. Nevertheless, the case demonstrates why technical capacity to produce an image does not, by itself, establish legal authorship or standing to enforce copyright.

Case 10. Thaler v. Commissioner of Patents — Australia

This case is sometimes confused with the later High Court decision. The earlier Federal Court decision, Thaler v Commissioner of Patents 20212021 FCA 879, initially accepted that an AI system could be an inventor under the Australian legislation as then interpreted.

That ruling was subsequently overturned by the Full Court of the Federal Court in Commissioner of Patents v Thaler 20222022 FCAFC 62, and the High Court later dismissed the appeal.

Importance: The sequence illustrates how judicial interpretation can change as appellate courts consider statutory language and the legal meaning of inventorship. The earlier first-instance ruling should not be cited as the current governing Australian position.

6. Comparative analysis of the cases

The cases demonstrate that AI intellectual property law is not governed by a single universal principle.

The cases support three broad conclusions.

First, ownership of AI technology does not automatically confer ownership of every output. Ownership must arise under the relevant copyright, patent, contract, employment, or assignment rules.

Second, AI training and AI-generated outputs present separate legal issues. A model might produce a non-infringing output even though its training process is challenged, or it might produce an infringing output despite the legality of its training.

Third, human creative contribution remains a key factor in many jurisdictions, but its legal significance varies according to the statute and the type of IP right claimed.

7. Major legal challenges in AI intellectual property governance

7.1 Ownership of AI-generated works

A major problem is determining who should own a work when multiple participants contribute to its creation.

For example, a generative system may involve:

A developer who designs the model.

A company that trains and deploys the system.

A user who provides prompts and reference materials.

An editor who modifies and selects the output.

A business that commissions the resulting work.

These participants do not necessarily acquire the same rights. The developer may own copyright in original software code, while the user may own copyright in qualifying human-authored additions to an AI-generated image. Contractual terms may allocate rights between the parties, but a contract cannot automatically create statutory copyright in material that does not qualify for protection.

7.2 Copyright infringement in training data

AI models often learn from large collections of text, images, music, videos, software, and other materials.

The legal problem is whether the acts required to collect and process those materials fall within exclusive copyright rights or an applicable exception.

A proper analysis should distinguish:

Accessing and collecting the work.

Making copies during storage or preprocessing.

Using the work for training or computational analysis.

Retaining information in model parameters.

Generating output that reproduces protected expression.

Distributing or commercially exploiting the output.

Different jurisdictions may treat these activities differently. An exception permitting text and data mining does not necessarily authorise every subsequent use or the production of infringing outputs.

7.3 The originality problem

Traditional copyright law protects original expression rather than ideas in the abstract.

Generative AI complicates this distinction because a system may produce numerous outputs in response to similar prompts. A user's contribution may consist of general instructions, highly specific instructions, repeated experimentation, or extensive post-generation editing.

The key question is whether the person's contribution meets the applicable legal standard for authorship and originality.

For example, a person who enters “create a landscape at sunset” and accepts the first output may have a weaker authorship claim than someone who makes detailed expressive choices, combines multiple outputs, and substantially edits the final composition. This is an illustration, not a universal legal test.

7.4 Patent inventorship and AI-assisted innovation

AI systems can assist in identifying chemical compounds, designing mechanical structures, generating technical solutions, and analysing engineering problems.

However, patent law must determine who qualifies as the inventor. A system's technical contribution is not equivalent to legal inventorship.

The challenge is to preserve meaningful patent protection for human-led innovation without automatically granting exclusive rights to machine-generated solutions in circumstances that do not satisfy statutory requirements.

7.5 Trade secrets and confidential information

AI developers may rely on confidential datasets, model parameters, training procedures, and proprietary software. These may be protected by trade-secret law where the applicable requirements are met.

At the same time, employees and users may inadvertently disclose confidential material by submitting it to an external AI system. Governance should therefore address access controls, confidentiality agreements, retention policies, and the use of submitted information for further training.

7.6 Enforcement and jurisdiction

AI systems can be trained in one country, hosted in another, and used to generate outputs for users worldwide.

This creates questions concerning territorial copyright law, applicable exceptions, jurisdiction, evidence, and remedies.

The Getty Images v Stability AI litigation illustrates the importance of establishing where the legally relevant conduct occurred. A claimant cannot assume that every alleged training activity is governed by the law of the country in which an output is later accessed.

8. What should an effective comparative governance framework contain?

A balanced governance framework should combine statutory clarity, technological transparency, contractual safeguards, and judicial oversight.

1. Clear rules on human authorship

Define how human creative contribution is assessed, including the role of selection, arrangement, editing, and expressive control.

2. Training-data accountability

Require appropriate records of data provenance, permissions, applicable exceptions, and rights reservations. Transparency should be balanced against privacy, security, and legitimate trade secrets.

3. Proportionate liability

Distinguish the responsibilities of developers, deployers, and users according to their conduct, knowledge, control, and applicable legal duties.

4. Transparent registration and disclosure

Establish practical procedures for identifying human contributions in copyright applications and identifying inventors in patent applications.

5. Licensing and collective solutions

Encourage licensing arrangements, collective rights management, and negotiated access to datasets where appropriate, without assuming that every training use requires the same legal treatment.

6. International cooperation

Promote greater consistency in terminology, evidence standards, cross-border enforcement, and licensing practices while respecting differences between national legal systems.

9. Specific implications for India

India has a particularly important role in comparative AI governance because its existing intellectual property statutes already contain provisions relevant to computer-generated works, while generative AI creates questions that those provisions may not have anticipated in their present form.

9.1 Copyright Act, 1957

Section 2(d)(vi) defines the author of a computer-generated literary, dramatic, musical, or artistic work as the person who causes the work to be created.

This provision offers a statutory starting point for analysing AI-generated material. Nevertheless, applying it to contemporary generative AI requires careful consideration of the particular system, the person's contribution, the work's originality, and the scope of the relevant rights.

9.2 Patent Act, 1970

Patent applications must comply with statutory requirements concerning inventorship, entitlement, novelty, inventive step, and industrial applicability.

India should distinguish inventions produced with AI assistance from claims that AI itself holds legal inventor status. The latter question should be resolved through authoritative statutory interpretation or legislation rather than assumed from the technical capabilities of AI.

9.3 Licensing and training-data governance

India could develop clearer guidance concerning:

Licensed and unlicensed training datasets.

The application of existing copyright exceptions to machine learning.

The use of copyrighted works for commercial generative AI training.

The protection of artists, authors, performers, publishers, and software developers.

Disclosure and evidentiary requirements in infringement proceedings.

Licensing arrangements that enable lawful access to creative works.

A workable framework should provide legal certainty without creating unnecessary barriers to scientific research, education, software development, and legitimate innovation.

9.4 A balanced Indian approach

A suitable approach would preserve the existing statutory foundation while clarifying how it applies to modern AI systems. Such clarification could include rules on human creative contribution, data provenance, training exceptions, contractual allocation of rights, and remedies for demonstrable infringement.

The objective should not be to grant AI systems unrestricted ownership, nor to prohibit AI-assisted creation. It should be to allocate rights and responsibilities in a way that promotes innovation and protects legitimate human and commercial interests.

10. Conclusion

Civil law and comparative AI intellectual property governance reveal a fundamental tension between established concepts of human authorship and inventorship and the increasing technical capacity of AI systems to produce creative and inventive outputs.

The European Union emphasises copyright compliance and regulatory transparency. The United States places substantial weight on human authorship and case-specific infringement analysis. The United Kingdom combines conventional copyright principles with a specific rule for computer-generated works. India has a statutory provision relevant to computer-generated works but faces unresolved questions about its application to modern generative AI. China has recognised human-directed AI output in a particular copyright dispute, while Japan provides a comparatively flexible framework for certain analytical uses of protected material.

The cases involving Thaler, Zarya of the Dawn, the Beijing Internet Court, Andersen, Getty Images, and Naruto illustrate the different legal dimensions of this problem. They show why authorship, inventorship, training, ownership, infringement, and enforcement must be examined independently.

The most defensible governance model is one that protects genuine human creativity, provides lawful routes for AI training, establishes clear rules for AI-assisted inventions, and imposes proportionate accountability for infringement. Comparative legal analysis is essential because no single jurisdiction has resolved every question raised by generative AI.

Academic note: This discussion is intended for educational and research purposes. It summarises selected statutes and judicial decisions, not legal advice. AI copyright and patent law continue to develop, and the legal effect of a decision depends on its jurisdiction, procedural status, and specific facts.