RETHINKING AUTHORSHIP AND LIABILITY IN THE AGE OF
GENERATIVE AI: A LEGAL ANALYSIS OF COPYRIGHT
CHALLENGES AND THE ROAD TO REFORM
ABSTRACT
The rise of Artificial Intelligence (AI), particularly generative AI, is transforming the
operations of many professions, including the legal sector. AI tools are now enabling lawyers
and students to enhance their efficiency by streamlining tasks such as summarizing court
judgments or drafting case briefs, making these processes faster and simpler. Nevertheless,
despite its advantages, AI presents complex challenges related to copyright and ownership.
This paper examines the legal and ethical dilemmas associated with AI-generated content,
focusing on issues of originality and the risk of inadvertently reproducing copyrighted
material.
This paper employs a qualitative approach to compare the methodologies adopted by various
nations, including India, the United States (US), the European Union (EU), and China, in
addressing copyright challenges associated with artificial intelligence (AI). A thorough
examination of Indian copyright law is undertaken through a discussion of significant cases
such as RG Anand v. Deluxe Films and Ram Sampath v. Rajesh Roshan, which elucidates the
interpretation of originality and fair use in India in accordance with the Copyright Act of
1957.
This paper aims to address a central inquiry: Who should bear the responsibility if artificial
intelligence infringes upon copyright laws? Should this liability fall upon the individual who
utilized the AI, the creator of the AI, or the AI itself? Given that current Indian legislation
recognizes only humans as authors, this paper contends that, under the present framework, AI
cannot be held legally accountable at this time. Nevertheless, as AI technology advances,
there may be a need to reconsider this stance. By examining the concepts of corporate
personhood and rights for non-human entities, this paper posits that the prospect of granting
AI a formal legal status warrants consideration in the near future.
Ultimately, the paper posits that Indian Copyright laws ought to be revised to effectively
address the emerging challenges and opportunities presented by artificial intelligence.
Page | 1
AI’S ROLE IN CREATING LEGAL CREATION AND COPYRIGHT
INFRINGEMENT
Artificial Intelligence (AI) has become a significant component of our daily lives. Whether
one is reading judgment summaries, drafting case briefs, or conducting various legal research
activities, it resembles having an intelligent assistant capable of operating continuously
without the need for rest. AI is transforming our methods of operation, simplifying and
streamlining processes. However, despite the excitement and potential it presents—
particularly for students and lawyers who face extensive reading and writing requirements—
it also raises critical Concerns arise regarding the ownership of the generated content and the
associated implications for copyright.
Consider a scenario in which artificial intelligence generates a legal summary. It may
assimilate information from previous case law and scholarly texts. If I were to present that
output as my own creation, would it be recognized as my original work? Furthermore, what
are the implications if AI-generated content incorporates sentences from copyrighted
materials? Does this constitute plagiarism, or is it permissible since it is categorized as AI
'creation'? Such inquiries have recently become prominent issues. Numerous writers and
artists express their concerns regarding the utilization of their works to train artificial
intelligence without their knowledge or consent.
For instance, during the 2022 Colorado State Fair1, Jason M. Allen, an accomplished game
designer, was awarded first prize in a digital art contest for a piece created using the AI tool
Midjourney. While he did make specific modifications, the primary creative process was
carried out by the AI. Numerous artists contended that utilizing AI to triumph in a human art
competition was unjust. Some likened it to “running a marathon in a self-driving car." The
judges were unaware of the AI-generated origin and asserted that they would have awarded
him the prize based solely on the artistic merit. This incident has now become a pivotal issue
in the ongoing copyright debate. When Allen subsequently attempted to copyright the image,
his request was denied, prompting him to file a lawsuit against the U.S. Copyright Office 2.
1
Cade Metz, 'An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy.' (2 September 2022)
[Link] accessed 20 May 2025.
2
Blake Brittain, 'US Copyright Office Denies Protection for Another AI-Created Image' (6 September 2023)
[Link]
Page | 2
This situation illustrates the increasing use of AI in competitive settings, igniting discussions
regarding fairness, authorship, and the essence of human creativity.
Copyright law provides legal protection for original creations. Over time, this law has
adapted to include rights for creating adaptations, derivative works, display, and performance,
emphasizing the use of intellectual property in various contexts. The situation becomes
complex with AI, as it is trained on vast amounts of data, which frequently includes
copyrighted creative works3. For AI to effectively conduct legal research, it requires training
on thousands of documents such as law reports, case summaries, and expert legal
commentaries, many of which are copyrighted. This raises concerns because these original
works are often utilized without the consent of the owners or authors, thereby violating their
exclusive rights under copyright law4.
Conversely, artificial intelligence is capable of generating content that is significantly akin to
an author's original work, thereby resulting in the reproduction of the original work 5. This
situation has engendered apprehensions regarding copyright infringement in both scenarios,
namely, the manner in which AI is trained (the input) and the nature of what it produces (the
output). A principal dilemma that arises in this context is the opaque nature of AI systems,
wherein users and, in some instances, even the developers lack complete understanding of
how the AI produces outputs or the reasons behind the resemblance to protected works in
certain outputs. This circumstance prompts further inquiries regarding whether the observed
similarity is unintentional or merely incidental.
For instance, Luke, a young man, uses AI to generate a romantic song for his girlfriend by
giving an innocent prompt. Neither he nor the AI developer intended to infringe copyright,
but later, it came out that the song was strikingly similar to Ed Sheeran’s. Should Luke be
liable? Or should the developer be liable? Or should the AI model, which used its internal
black box6 mechanism to generate the output, be held liable?
As legal scholar Jane C. Ginsburg once said, “Copyright law’s purpose is not to reward the
labor of authors but to encourage the creation and dissemination of their work 7,” which
09-06/ accessed 26 May 2025.
3
Adam Buick, 'Copyright and AI Training Data—Transparency to the Rescue?' (2025) 20 JIPLP 182.
4
Daniella Smith, Growing Your Library Career with Social Media (1st edn, Chandos Publishing 2018).
5
Matt Blaszczyk, Geoffrey McGovern, and Karlyn D. Stanley, Artificial Intelligence Impacts on Copyright Law
(20 November 2024) [Link] accessed 26 May 2025.
6
Matthew Kosinski, 'What Is Black Box AI and How Does It Work?' (29 October 2024)
[Link] accessed 26 May 2025.
7
Jane C Ginsburg, 'The Role of the Author in Copyright' in Ruth L Okediji (ed), Copyright Law in an Age of
Limitations and Exceptions (Cambridge University Press 2017).
Page | 3
means copyright is not only about ownership but also about protecting creativity and
encouraging people to keep creating. If AI can freely use someone’s creative content without
permission or payment, it could discourage future creators and hurt the legal content industry.
Another challenge that emerges is understanding how AI works in reality. Copyright
infringement leads to the unauthorized exploitation of creators' intellectual efforts due to the
unlicensed use of their work, for which they are not being rewarded 8. To restore the infringed
rights of the creators, the most viable way is to compensate them for the damages incurred by
them.9 The person liable needs to be compensated, but it is hard to hold an entity liable in this
complex process from input to output.
This raises an imperative question: Who is responsible for copyright infringement? Is it the
developer who made the AI? Or the user? Or someone else? So far, there is no mechanism to
judge liability in the case of copyright infringement caused by artificial intelligence.
Especially in a situation like Luke’s case, where neither intent nor awareness existed, yet the
output still resembles an existing copyrighted work, deciding responsibility becomes
increasingly complex and unclear.
METHODOLOGICAL APPROACH
This paper is qualitative research that combines literature analysis, comparative analysis, and
legal interpretation methods. This paper tries to understand the latest developments in the
realm of copyright due to the coming of generative AI by looking at the intricate elements
and theories guiding the functioning of copyright law worldwide. This paper compares and
contrasts the new legislative additions made in the US, UK, EU, and China regulations on the
use of AI to understand the vision of leading, developing, and developed nations. This paper
uses the Indian jurisprudence developed by the decisions of the courts of law to interpret the
Indian Copyright Act concerning the recent developments in technology in use, generating
more avenues for the infringement of original work. The central focus of this paper is to
analyze the parties' liability for the novel form of copyright infringement, determining their
liability by establishing ownership over the output by the generative by backing it with legal
bases. The aim is that this examination will contribute to the equitable resolution of the
copyright infringement liability arising from AI-generated content and suggest some policy
8
S.S. Rana & Co., 'Copyright Misuse and Infringement in India' (2023) [Link]
india/copyright-misuse-and-infringement-india/ accessed 26 May 2025
9
ClearTax, 'Copyright Infringement: Meaning, Examples, Cases in India' (2025) [Link]
infringement accessed 26 May 2025.
Page | 4
changes and interventions to make the Indian Copyright Act better equipped to tackle the new
genre of challenges arising from generative AI in the field of copyright.
AI OUTPUT INFRINGEMENT: LEGAL TEST AND COMPARATIVE
COPYRIGHT STANDARDS
Copyright Act globally and in India under Section 13 10 of the Copyright Act protects the
expression of an idea, not the idea itself. This distinction is crucial in determining
infringement. If AI output shares the same concept as an existing work but with a different
expression, it doesn’t count as copyright infringement. However, if the output shares almost
the same form, structure, or creativity as the original, it might fall under copyright
infringement.
In the case of R.G. Anand v. Deluxe Films & Ors. 11, held in the year 1978, the Supreme Court
mentioned that having a similar idea or theme is insufficient to prove copyright infringement.
The main thing that matters is how the concept has been expressed. The court mentioned that
the primary test is whether an average person, with standard memory and understanding,
would look at both works and feel like no one is a copy of the other. This is known as the
Ordinary Observer Test12. It does not focus on small technical details but instead looks at the
overall feel or impression of the work. If the new job comes out to be very similar to the
original, especially in the unique and creative parts, it could still be copied, even if some
other parts are different.
In the Ram Sampath v. Rajesh Roshan case 13, the Bombay High Court held that even a short
music phrase like the catchy part of a tune called “hook” could be enough to amount to
infringement, especially if people instantly recognize it. Similarly, Twentieth Century Fox v.
Sohali Maklai Entertainment Pvt. Ltd. & Anr14 held that key creative elements, even in small
part, violate copyright if those parts represent the creator’s creativity.
Under section 14 of the Indian Copyright Act15 talks about adaptation, i.e., converting a work
to another format. Sometimes, people change jobs instead of copying them directly; that is
10
Copyright Act 1957 (India), s 13.
11
R.G. Anand v. M/S Deluxe Films & Ors [1978] AIR 1613 (SC).
12
Christopher V Carani, 'The New "Extra-Ordinary" Observer Test for Design Patent Infringement—On a
Crash Course with the Supreme Court's Precedent in Gorham v. White' (2009) 8 J Marshall Rev Intell Prop L
354.
13
Ram Sampath v. Rajesh Roshan & Ors. [2008] SCC Online Bom 370 (Bom HC).
14
Twentieth Century Fox Film Corporation v. Sohail Maklai Entertainment Pvt. Ltd. & Anr. [2010] 44 PTC 647
(Bom).
15
Copyright Act 1957 (India), s 14.
Page | 5
what is called an adaptation. In 1994, it was held that even if an altered or restructured
version exists that feels like the original, it can be counted as an adaptation.
In Barbara Tylor Bradford v. Sahara Media Entertainment Ltd. and Ors 16, the court restricted
the scope of the word “alter.” It said that minor alterations that do not change the work or the
purpose of the original work aren’t sufficient to avoid the copyright rule. So, if AI generates
an output that doesn’t change the core or primary purpose of the protected work, that can also
result in a copyright issue.
Under section 52 of the Indian Copyright Act17, India recognizes fair dealing for purposes like
research, criticism, or education. That might be allowed if we instruct AI to write a research
project and include a small quote from a book. But if AI generates content that looks like the
other creator's protected work and gets published or shared among different people, that
wouldn’t be protected.
In context with the European Union18, there is a strict set of rules known as the EU AI Act. It
does not deal directly with copyright issues. Instead, it also includes rules for how AI systems
should behave, primarily when used for high-risk tasks. Under the Digital Single Market
Directive, AI can only do text and data mining(TDM) in some situations, especially while
dealing with copyright-protected material19. This proves that the EU is also trying to ensure
AI follows proper copyright laws, especially when it creates content.
However, the U.S. has a more flexible fair use system, 20 which can allow commercial use in
some cases, especially if the newly generated work is unique. At the same time, India’s laws
contrast and don’t offer as much room for interpretation.21
In contrast, the European Union’s rules are more structured. They don’t just observe how AI
is trained but also how it is used. Their method combines AI regulation with copyright law,
meaning that developers are users of AI tools and need to be careful not to break IP rules.
16
Barbara Taylor Bradford & Anr. v. Sahara Media Entertainment Ltd. & Ors. [2003] 2004 (28) PTC 474
(Cal).
17
Copyright Act 1957 (India), s 52.
18
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down
harmonised rules on artificial intelligence and amending certain Union legislative acts (Artificial Intelligence
Act) [2024] OJ L 2024/1689.
19
Adam Buick, 'Copyright and AI Training Data—Transparency to the Rescue?' (2025) 20 JIPLP 182
[Link] accessed 26 May 2025.
20
Stanford Copyright and Fair Use Center, Measuring Fair Use: The Four Factors (12 December 2012)
[Link] accessed 26 May 2025.
21
Campbell v Acuff-Rose Music Inc 510 US 569 (1994).
Page | 6
Such a system could be a good example for India if we ever want to create a special legal
framework for AI and copyright.
AI models are usually not designed to copy directly from what they were trained. 22 However,
this may happen in rare cases, mainly if someone uses tricks like prompt injection, 23 in which
users deliberately try to give a prompt, by which an AI might generate content that is very
close to the original content.
There are two key scenarios of concern:
1—Verbatim reproduction—Verbatim is a Latin word that means “word for word.” This
happens when the AI generates almost the same output of the data it was trained in, like a
paragraph from a book or a commentary.
2—Minor alteration happens when AI generates small surface-level changes, like rephrasing
sentences or altering a few words. Still, the overall structure and purpose of the original
content remain almost the same. As held in Ram Sampath v. Rajesh Roshan, copying even a
distinctive phrase may amount to infringement if it captures the creativity of the original
work.
Either case challenges authorship by creating a ‘shadow’ of the original protected content; it
just generates the same work with slight new [Link], It is concluded that if an output
is generated that substantially resembles the original work then it will have an infringing
falling under no exceptions provided by the contract act. Thus leading to the discussion of
who should be held liable for this infringement.
In contrast, the European Union’s rules are more structured. They don’t just observe how AI
is trained but also how it is used. Their method combines AI regulation with copyright law,
meaning that developers are users of AI tools and need to be careful not to break IP rules.
Such a system could be a good example for India if we ever want to create a special legal
framework for AI and copyright.
22
Rita Matulionyte, ‘Reconceptualizing the Reproduction Right in the Age of AI’ (2025) Kluwer Copyright Blog
[Link]
accessed 26 May 2025.
23
Matthew Kosinski and Amber Forrest, ‘What Is a Prompt Injection Attack?’ (IBM, 26 March 2024)
[Link] accessed 26 May 2025.
Page | 7
DETERMINING AUTHORSHIP IN AI-GENERATED WORKS: LEGAL
OWNERSHIP AND LIABILITY PERSPECTIVE
This paper endeavors to elucidate the notion of liability by delineating the authorship or
ownership of content generated by artificial intelligence under the tenets of direct liability as
upheld in Indian jurisprudence. The guidelines, anchored in the foundational principles of
strict liability,24 assert that the individual responsible for generating the output and possessing
ownership thereof has directly violated the copyright owner's rights through such actions,
thereby rendering the owner liable. Within the framework of copyright law, it is relatively
commonplace for a specific protocol to impose liability upon an individual who reproduces
an original work, as mandated by section 51 of the Indian Copyright Act (ICA), 25 which
unequivocally states that infringement occurs when an individual exercises any exclusive
rights inherent to the original author without having obtained authorization to do so.
To comprehend the concept of ownership, it is essential first to identify and assess all relevant
parties that may assert ownership over the produced output. The initial entity could be
artificial intelligence that utilizes the information upon which it is trained to generate
statistical predictions and yield results. The second entity, potentially the author, refers to the
AI developer who trained the AI using the pertinent datasets that enabled it to produce
outputs. The third-party contending for ownership is the user who provided valuable
commands that facilitated the AI in producing unique output. Lastly, the fourth relevant
participant could be the service provider equipped with AI at its site, enabling the user to
interact and achieve results and treating the AI as a work made for hire. The paper deals with
each contested party individually.
THE ARTIFICIAL INTELLIGENCE AS A CREATOR: LEGAL
BARRIERS TO AUTHORSHIP AND PERSONHOOD
Artificial Intelligence (AI) is a unique creator, generating output through a complex process
that involves data analysis, storage, knowledge extraction, and various similar functions. 26
The standard definition used to characterize generative AI refers to content created from the
24
Legal Information Institute, Strict Liability (October 2024) [Link]
accessed 26 May 2025.
25
Copyright Act 1957 (India), s 51.
26
Usama Fayyad, Gregory Piatetsky-Shapiro and Padhraic Smyth, ‘From Data Mining to Knowledge
Discovery in Databases’ (1996) 17 AI Magazine 37.
Page | 8
ground up using a specific machine learning algorithm. 27 These algorithms develop a model
from which the AI learns, applying mathematical representations that emulate real-world
learning processes through data sets. 28 This learning process empowers the AI to make
statistical predictions, producing novel data samples that resemble the original data it was
initially trained on. It is crucial to acknowledge that, by the intrinsic nature of AI
programming, it is designed to produce outputs that resemble the original rather than directly
replicating it.29 The AI mimics a memorization process to generate outputs, which involves
remembering the basic outline and capturing the essence by classifying data points. The
outputs are not derivatives of any specific work within the datasets; this is why many datasets
are used. However, some outputs substantially similar to a particular data point can be formed
because there may be many data points regarding a specific subject. 30 For instance, if there
are a million data points in a dataset based on superhero characters, and a majority of them
feature Superman, then it is evident that an AI trained on such a dataset will generate
Superman when prompted to create a superhero, which could infringe the copyright of
Superman's owner.
The workings of AI suggest that, like human beings, they learn from the experiences and
content circulated and collected from their surroundings and, based on that understanding,
produce an output that expresses their thoughts and understanding. 31 The workings of an AI
suggest that its outputs have a character of originality ingrained in them, 32 but that might not
be the case in real life. The key element for granting a copyright is the originality of the
expression. Some may argue that the AI works based on prompts given by the user, and
therefore, it is not the creation of the AI itself. Then, it becomes clear to define the
relationship between the AI and the user, determining if it acts as a tool, an independent
27
Chiara Longoni and others, ‘News from Generative Artificial Intelligence Is Believed Less’ (2022) ACM
Conference on Fairness, Accountability, and Transparency (Association for Computing Machinery 2022)
[Link] accessed 25 May 2025.
28
Yang Lu, ‘Artificial Intelligence: A Survey on Evolution, Models, Applications and Future Trends’ (2019) 6
Journal of Management Analytics 1.
29
Andrés Guadamuz, ‘A Scanner Darkly: Copyright Liability and Exceptions in Artificial Intelligence Inputs
and Outputs’ (2024) 73 GRUR International 111.
30
Andrés Guadamuz, ‘A Scanner Darkly: Copyright Liability and Exceptions in Artificial Intelligence Inputs
and Outputs’ (2024) 73 GRUR International 111.
31
Stephen Wolfram, ‘What Is ChatGPT Doing … and Why Does It Work?’ (Stephen Wolfram Writings, 14
February 2023) [Link] 25
May 2025.
32
Yogesh K Dwivedi and others, ‘So What if ChatGPT Wrote It? Multidisciplinary Perspectives on
Opportunities, Challenges and Implications of Generative Conversational AI for Research, Practice and Policy’
(2023) 71 International Journal of Information Management 102642.
Copyright Act 1957 (India), s 13(1)(a).
Page | 9
contractor, or if the work qualifies as work for hire, which will be discussed in the following
sections. The user's prompts can be treated as delivering ideas that cannot be copyrighted.
The originality of the content is determined as specified in section 13 1a) of the ICA, 33 which
applies to original literary, dramatic, musical, and artistic works created by individuals
throughout India. This use of the term originality has not been defined in the ICA, to which
many doctrines have been applied, considering the relevant approach used in other
jurisdictions. In the jurisdiction of the UK, the sweat of the brow doctrine was used, holding a
light hand on the creativity behind the work while granting it the title of originality. 34 This
doctrine emphasized the role of effort, labor, and skill in creating rather than the creativity
behind it as long as it originated from the creator. If this doctrine is applied, the copyright
rights of the AI over its generated output may seem far-fetched.
According to this ancient doctrine, the immediate results and the ease with which the AI
produces them would not meet the required threshold. Another doctrine used in the US was a
modicum of creativity,35 which required the work to demonstrate at least a minimal level of
creativity to qualify for copyright. This copyright demands more from the work than mere
skill or effort; even a pinch of creativity meets the requirement. The Indian jurisprudence
adopted a midway between the two by the Indian SC in the Eastern Book Company & Ors v.
D.B. Modak & Anr.36 The skill and judgment test developed by the Canadian SC 37 proposes
to assess the work due to the author's ability to use knowledge and discernment. The Indian
courts demand that a creation reflect effort, creativity, and intellectual contribution. 38 This
combination goes against the proposition of AI being treated as a suitable writer. It could put
to rest the questions of creativity and the use of an inferred intellectual process by analyzing
the process of creation with a tint of modern understanding overseeing it. The issue of labor
and sweat strikes a chord, becoming a hindrance in establishing the competence of AI as an
author. With the growing time, this might change a modern interpretation, molding it to the
needs of the time.
33
Copyright Act 1957 (India), s 13(1)(a).
34
J K Das, ‘Copyright Protection for Computer Softwares and Databases: From Sweat of the Brow to Modicum
of Creativity’ (2012) 3 Indian Journal of Law and Justice 57.
35
Howard B Abrams, ‘Originality and Creativity in Copyright Law’ (1992) 55 Law & Contemporary Problems
3.
36
Eastern Book Company v D B Modak AIR 2008 SC 809.
37
CCH Canadian Ltd v Law Society of Upper Canada, 1 SCR 339.
38
Tanya Saraswat, ‘Evolution of Tests of Creativity in Copyrights’ (IP and Legal Filings, 2023)
[Link] accessed 26 May 2025.
Page | 10
Even if this is set aside, the pertinent question of personhood has yet to be decided, as Section
2(d)(vi) of the ICA39 indicates that the author who would receive exclusive rights for their
work must be a natural person. This raises significant legal implications and further
complicates the issue of AI ownership under Indian copyright law.
Some scholars support the idea that an author must be human based on the incentive theory, 40
the foundation of copyright laws. This theory suggests that copyright should be granted to
works created by authors to encourage others to produce original, creative, and artistic works,
benefiting society. An AI, being an artificial entity, does not require any incentive to generate
work. In the US, the copyright office 41 has closed the door to recognizing AI as an
independent author of the works it produces. However, there is still some speculation about
copyright in works created by AI that involve a significant amount of human input. Many
prestigious journals, such as Nature,42 have altered their authorship guidelines to prevent AI
from being considered authors because authors should be accountable for their work.
In contrast, AI currently lacks a system for establishing liability. Authorities may hold AI
accountable for its output, but this would lead to no consequences due to the lack of property
associated with AI.43 Even in copyright infringement cases, the best course of action for a
victim is to file a civil suit to recover damages caused by the AI's infringing output. However,
AI has no property to set aside, leaving the victim as dented as before.
Although Indian Copyright law clearly states that only a human being can be the author of a
work, it is interesting to view how the law treats other non-human entities. For example,
companies and animals have, in some cases, been given legal rights or personhood. A
company is not a human being; it can still own property, sue, and be sued. Likewise, in some
parts of the world, animals have received certain legal protections not just because they are
living beings but because the law recognizes the importance of them in some form of
recognition.44 This raises the question: if these non-human entities can be regarded as legal
39
Indian Contract Act 1872, s 2(d)(vi).
40
Diane Leenheer Zimmerman, ‘Copyrights as Incentives: Did We Just Imagine That?’ (2011) 12 Theoretical
Inquiries in Law 29.
41
Hafiz Gaffar and Saleh Al-Barashdi, ‘Copyright Protection for AI-Generated Works: Exploring Originality
and Ownership in a Digital Landscape’ (2024) Asian Journal of International Law
[Link] accessed 26 May 2025.
42
Nature Portfolio, ‘Artificial Intelligence (AI) – Editorial Policies’ (Nature, 2024)
[Link] accessed 26 May 2025.
43
Yuyan Yang, ‘Attribution of Liability for Copyright Infringement by Artificial Intelligence Generated Content’
(2023) Lecture Notes in Education Psychology and Public Media [Link]
7048/29/20231424 accessed 26 May 2025.
44
Jiahong Chen and Paul Burgess, ‘The Boundaries of Legal Personhood: How Spontaneous Intelligence Can
Problematise Differences Between Humans, Artificial Intelligence, Companies and Animals’ (2019) 27 Artificial
Page | 11
individuals in certain situations, then why not AI? AI indeed does not have feelings, but
companies also don't have the same. AI sometimes creates things on its own, just like a
person would. So, instead of thinking of AI as just a tool, the law should start thinking about
AI differently as it is becoming more advanced daily. This does not mean AI should
immediately get its rights like a person, but it shows that the idea is not as unrealistic as it
might look.
This suggests that AI cannot be a relevant author or a liable entity at its current stage, but
upcoming developments in the realm of AI will definitely require changes in how we
perceive AI's authorship and personhood. Until that day comes, AI cannot be considered to be
the appropriate owner of the work created and, therefore, cannot be held liable for the
infringement caused by the work thereof.
THE USER AS AUTHOR: ASSESSING HUMAN CONTRIBUTION IN
AI-GENERATED WORKS
The user plays a crucial role in the legal dispute regarding ownership of the output generated
by the user's prompt. Their contribution to creating the work aligns with the originality
principles used to evaluate the uniqueness of creations. The user plays a crucial role in the
legal dispute regarding ownership of the output generated by the user's prompt. Their
contribution to creating the work aligns with the originality principles used to evaluate the
uniqueness of creations. Some scholars argue45 that the user exercised creativity through the
complexities introduced in the prompt, enabling the AI to produce results with specific
characteristics in creative expression. The underlying idea behind this argument is to use AI
as a tool. Consider the example of a photographer who employs understanding, skill, and
judgment to take a photograph with a camera. This process may involve factors such as
brightness, sharpness, and exposure from the camera. However, this does not diminish the
photographer's skill in capturing images while being aware of surrounding elements and
expressing creativity. This process may involve factors such as brightness, sharpness, and
exposure from the camera. However, this does not diminish the photographer's skill in
capturing images while being aware of surrounding elements and expressing creativity. One
cannot argue that the photographer merely clicked the button on the camera, overlooking the
Intelligence and Law 73 [Link] accessed 26 May 2025.
45
Jonas Oppenlaender, Rhema Linder and Johanna Silvennoinen, ‘Prompting AI Art: An Investigation into the
Creative Skill of Prompt Engineering’ (2024) International Journal of Human–Computer Interaction
[Link] accessed 26 May 2025.
Page | 12
effort and labor essential to the doctrine of the sweat of the brow. This case differs when AI
utilizes its black box mechanism 46to generate results, making it difficult for a person to assess
the adequate input provided in producing the output. Granting copyright for production with a
high degree of machine-determined expressive elements would violate the fundamental
principles of rewarding the author for their labor and acknowledging the creation as a product
of their mind. This view must also consider the perspective of authors 47 who believe that AI
needs to interact with users to generate output. Another aspect of creativity may raise
questions about the user's originality in the production, as the AI ultimately created the
expression and infused some creative elements into the result. Some scholars have suggested
a test for human involvement related to the product. 48 If the product was generated from a
highly descriptive prompt, then the AI's contribution to the work might be minimal, treating it
more as a tool than a co-creator. According to the US Copyright Office, 49 copyright would be
granted to the user if they use sufficient expressive elements such that a sense of human-
authored work can be perceived from the AI-generated output or if a human makes certain
modifications to the AI-generated output. The US Copyright Office does not consider the
mere provisions given to the prompt to hold any value as an introduction of human
expressive elements.
While delving into the authorship claims, one must also take care of the causation behind the
work that led to its existence. 50 These similar arguments were posed after the invention of the
camera on the authorship of the photograph, arguing that the picture was not creative and the
camera was the creator of the photograph, not the person behind it. 51 The photographer
recognized which photos best expressed the idea in his mind. The user inspires and guides the
AI to express his mind's desires, even though the AI takes steps to do so. The result may or
46
Matthew Kosinski, ‘What Is Black Box AI and How Does It Work?’ (IBM, 29 October 2024)
[Link] accessed 26 May 2025.
47
Dan L Burk, ‘Thirty-Six Views of Copyright Authorship, by Jackson Pollock’ (2020) 58 Houston Law Review
263 [Link]
accessed 26 May 2025.
48
Harshal Chhabra and Kanishk Gaurav Pandey, ‘Balancing Indian Copyright Law with AI-Generated
Content: The “Significant Human Input” Approach’ (IJLT, 26 February 2024)
[Link]
approach accessed 26 May 2025.
49
U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (29 January 2025)
[Link] accessed
26 May 2025.
50
Mackenzie Caldwell, ‘What Is an “Author”?—Copyright Authorship of AI Art Through a Philosophical Lens’
(2023) 61 Houston Law Review 411 [Link]
authorship-of-ai-art-through-a-philosophical-lens accessed 26 May 2025.
51
Justin Hughes, ‘The Photographer’s Copyright — Photograph as Art, Photograph as Database’ (2012) 25
Harvard Journal of Law & Technology
339 [Link] accessed 26 May 2025.
Page | 13
may not be the expression that the user wanted to express, but it delves into the description,
making the machine understand the form of the desired result. This human involvement is the
element that leads to an expression that has enough human elements. In the court's decisions,
the most important thing to remember is that AI is the machine allowing the intellectual work
of the user to take a form. It is evident from the above discussion that the user is the ultimate
owner of the AI-generated work. Extending copyright to users becomes the most practical
solution.
THE AI DEVELOPER’S ROLE IN COPYRIGHT AUTHORSHIP
The AI developer could contend over ownership, treating the AI as a work for hire, as
recognized in the US.52 The developer could easily attribute the AI as its employee, viewing
itself as an independent contractor hired by the user to perform a task. Although the task
result may bear some influence and inspiration from the user, the expression reflects the
creative coding and development invested in ensuring the AI functions effectively. When a
customer approaches a painter and asks for a portrait of himself with his wife, the result is a
painting of the customer, which he buys from the painter. The inspiration provided by the
customer does not entitle the customer to the copyright of the painting. The art is made from
the inspiration or idea given by the customer. However, the painting drawn is a product of the
artist's use of his artistic abilities, including the color contrasts, the use of a particular brush,
the shading, and more. The expression is the painter's product, and the paint is the painter's
tool, which helped him reach the result. This relationship between the customer and the
painter resembles that of an AI developer and the user. The copyright of the work created by
an employee under the apprenticeship or the contract of service will subsist in the employer
as stated in section 17c of the ICA. 53 The datasets and the necessary algorithms provided by
the AI developer for the creation of the AI itself grant the developer the right over the AI and
its results. The relationship between the AI, AI developer, and the user is of the utmost
importance in understanding the copyright over the generated output. The AI is treated as a
tool of the user or the employee or work for hire by the AI developer, which makes a vast
difference in establishing the copyright over the user and the AI developer.
This leads to the concept of co-authorship, which compensates the AI developer for the
algorithms, coding, and datasets vital for the AI's operation while rewarding the user for the
52
Rita Matulionyte and Jyh-An Lee, ‘Copyright in AI-Generated Works: Lessons from Recent Developments in
Patent Law’ (2022) 19 SCRIPTed 5 [Link]
recent-developments-in-patent-law/ accessed 26 May 2025.
53
section 17c of the ICA.
Page | 14
prompts provided, shaping the resulting output. This stage of co-authorship requires a
mechanism to fairly compensate both parties for their contributions and assign copyright
rights accordingly. Co-authorship can be established through a contract between the user and
the AI developer, outlining their responsibilities and rewards for the generated content. These
contracts, consisting of the terms and conditions 54 agreed upon by both parties, will facilitate
a mutual understanding of how copyrights are distributed, thereby resolving the disparities
stemming from ownership ambiguity.
TYPES OF LIABILITY FOR INFRINGEMENT IN AI-GENERATED
INFRINGEMENT: FROM DIRECT TO DUAL RESPONSIBILITY
Authorship is relevant to the discussion and has to be linked to the pertinent liabilities arising
from AI's infringing output. These liabilities will be put to work and help establish the
relevant authorities at fault. When someone infringes an author's copyright due to their
involvement in forming the infringing output through AI, different liabilities come into force
against them. The level and type of liability may vary based on the person's activity, how
much they were involved, and whether they intended to break the law. In India, the Copyright
Act of 1957 mainly focuses on direct infringement. Still, courts have also talked about other
kinds of liability, like contributory and vicarious, mostly while looking at how these issues
are handled in countries like the U.S. and the U.K.
1- Strict Liability (Direct Liability)
This is the primary category of liability. If someone directly does something that only the
copyright owner has the right to do, such as copying a book or uploading a movie, they are
directly liable for infringement.55 In India, there is no need to prove that the person meant to
break the law, In most cases, the mere act is enough.56 Copyright infringement is often treated
as a strict liability offense in India, 57 meaning a person can be held responsible even if they
didn't mean to break the law. The main idea is to protect creators and discourage careless use
54
The Indian Lawyer, The Importance of a Website’s Terms and Conditions and Privacy Policy (14 July 2016)
[Link] accessed 26 May 2025.
55
Pribanic & Pribanic, ‘Direct Liability Legal Definition’ (Pribanic & Pribanic, 2022)
[Link] accessed 26 May 2025.
56
PSL Advocates & Solicitors, ‘Limitation of Liability Clauses and Their Applicability in India: A Judicial
Perspective’ (Lexology, 13 February 2025) [Link]
4d60-bd67-17cfd76c1e02 accessed 26 May 2025.
57
Dipak Rao and Nishi Shabana, ‘Whether Infringement of Copyright Is a Cognizable Offence?’ (2015)
TerraLex Connections[Link]
%20Connections%220Whether%20Infringement%20of%20Copyright%20Is%20a%20Cognizable
%20Offence_.pdf accessed 26 May 2025.
Page | 15
of other people's work. Fair-dealing provisions help balance things out so that people using
content for education, research, or criticism are not punished unfairly.
For example, in the case of R.G. Anand v. Delux Films (1978), 58 the court observed how
much of a copyrighted work was copied in a film and decided that it was an infringement,
though there might not have been any bad intention involved. However, they may not be held
liable if the person can prove that their use falls under "fair dealing," as mentioned in Section
5259 of the Copyright Act.
2- Indirect or Contributory Liability
Sometimes, a person doesn't copy the work themselves while they help someone else do it.
This is called Indirect or Contributory Liability. 60 Indian law doesn't mention it directly, but
courts have started to look at it based on ideas similar to those of other countries.
To understand this, we can take the example of the 2017 Myspace v. Super Cassettes Case. 61
In this case, a user uploaded copyrighted songs on Myspace, and the company was held liable
for helping them do it. The court said that even if the company didn't upload the content
themselves, they knew what was happening and didn't stop, so they might still be responsible.
However, protection is being provided to internet platforms under the IT Act 62 if they act
quickly to remove content when asked.
3- Vicarious Liability
Vicarious Liability occurs when someone is found responsible for someone else's actions,
mainly if they benefit from them and have the power to control them. 63 For instance, if an
employer allows an employee to use pirated software at work and doesn't stop them from
doing so, the employer might be held liable.
58
R.G. Anand v. M/S Deluxe Films & Ors [1978] AIR 1613 (SC).
59
Copyright Act 1957, s 52.
60
Legal Information Institute, ‘Contributory Infringement’ (Wex, August 2022)
[Link] accessed 26 May 2025.
61
MySpace Inc v Super Cassettes Industries Ltd FAO(OS) (COMM) 21/2017, Delhi High Court, 23 November
2017.
62
Information Technology Act 2000, s 79.
63
Legal Information Institute, ‘Vicarious Liability’ (Wex, August 2022)
[Link] accessed 26 May 2025.
Page | 16
This idea has emerged in cases involving websites and app developers 64 who earn money
from copyrighted content users upload. Courts try to analyze whether the platform could have
controlled the situation and whether it earned money from it.
Upon understanding the liabilities that come into play above, it is imperative to grasp how
these are integrated within the dual co-authorship scenario in the context of output generated
through artificial intelligence. The nature of the infringement and the underlying reasons have
already been discussed in Part 1. Direct liability is established based on the specific role of
the user who provided particular instructions in the form of prompts, which led the AI to
generate results of an infringing nature. In Indian jurisprudence, no requirement exists to
establish intent to prove infringement; however, this situation leaves the user uncertain, as it
is impossible to predict how the AI will operate and the outcomes it will generate. This
scenario necessitates reform, as it contradicts the fairness principles inherent in the broader
ideals of justice. The user may have merely provided an innocuous prompt that resulted in an
infringing outcome; however, there are also instances where the user intentionally delivers a
special prompt that “hacks” the AI into producing reproductions of original works, as
contended by defendants in the cases of UMG Recordings v. Suno 65 and New York Times v.
Microsoft Corporation.66 This situation calls for establishing a mechanism to assess the
validity of the user-provided provisions, prompting the AI to yield such results.
The ordinary person test could be utilized to ascertain direct liability, specifically, to test
whether the provisions or instructions offered to the AI would lead an ordinary individual to
recognize an already available copyrighted work or not in answer to those instructions. If the
instructions lead an ordinary person to point to an original existing work that would satisfy
the result asked for, then the user would be liable. This test incorporates both the vagueness
and intent of the user. If a user asks for a work in the style of da Vinci, then the ordinary
person test could be used. If the ordinary person points to an already existing work of da
Vinci that would match the result and satisfy the command given, then the user would be held
liable.
The contributory liability of the AI developer will be universal for the infringement caused
by the AI. The AI developer is the one who used such datasets, which had in them the original
work of creators, which was taken without the permission of the relevant owner and used to
64
Prachi Tyagi, ‘Social Media and Copyright: An Indian Perspective’ (2023) 28 Journal of Intellectual Property
Rights 402 [Link] accessed 26 May 2025.
65
UMG Recordings, Inc. et al. v. Suno, Inc. et al., No. 1:24-cv-11611 (D Mass, filed 24 June 2024).
66
Page | 17
train the AI, empowering the AI to give reproductions resembling the work and hence
infringing the exclusive rights of the author. The AI developer could have used only non-
copyrighted work or licensed the use of the copyrighted work from the owner. Then, there
would have been no scope of liability upon him. The principles of attribution of liability also
suggest that correspondence between risks, benefits, and responsibilities should guide the
liability. Legal responsibilities should be allocated proportionately to the nature and scope of
risks each stakeholder (user and the AI developer) bears and the benefits they gain within the
generative AI ecosystem. The developer is the most suitable for bearing losses as he is the
one who charges for the services of the AI and, going by the vicarious liability principles, the
key person with the deep pockets to recover the damages of the victim. The Chinese court in
Shanghai Character License Administrative Co., Ltd. (SCLA) v. tab ( 67anonymity preserved)
held that there is a third-party due diligence on the service provider who uses AI on their site,
even though it did not create the generative AI itself. The Court’s findings that (i) similarity
between the output of a generative AI system and pre-existing works can amount to an
infringement of the rights in those pre-existing works, and (ii) the operator of a website
offering that an AI system can be liable for those outputs 68 guiding that due diligence has to
perform safeguarding the rights of others.
The paper proposes a system of dual liability with the AI developer as primary liable and the
user as a secondary liable party. After judging the user's intent, the secondary party could be
added to the proceedings based on the ordinary person test. The AI developer will be liable
under strict liability. There have been cases in which the AI developer develops a contract
with the user granting him the entire ownership over the content produced by the AI to escape
liability. Still, it is imperative to understand that if no original work had been used in the
training of AI, no such result would have ever been produced.
Upon understanding the liabilities that come into play above, it is imperative to grasp how
these are integrated within the dual co-authorship scenario in the context of output generated
through artificial intelligence. The nature of the infringement and the underlying reasons have
already been discussed in Part 1. Direct liability is established based on the specific role of
the user who provided particular instructions in the form of prompts, which led the AI to
generate results of an infringing nature. In Indian jurisprudence, no requirement exists to
67
Bird & Bird, ‘Liability of AI Service Providers for Copyright Infringement’ (2024)
[Link]
accessed 26 May 2025
68
.
Page | 18
establish intent to prove infringement; however, this situation leaves the user uncertain, as it
is impossible to predict how the AI will operate and the outcomes it will generate. This
scenario necessitates reform, as it contradicts the fairness principles inherent in the broader
ideals of justice. The user may have merely provided an innocuous prompt that resulted in an
infringing outcome; however, there are also instances where the user intentionally delivers a
special prompt that “hacks” the AI into producing reproductions of original works, as
contended by defendants in the cases of UMG Recordings v. Suno and Udio and New York
Times v. OpenAI. This situation calls for establishing a mechanism to assess the validity of
the user-provided provisions, prompting the AI to yield such results.
The ordinary person test could be utilized to ascertain direct liability, specifically, to test
whether the provisions or instructions offered to the AI would lead an ordinary individual to
recognize an already available copyrighted work or not in answer to those instructions. If the
instructions lead an ordinary person to point to an original existing work that would satisfy
the result asked for, then the user would be liable. This test incorporates both the vagueness
and intent of the user. If a user asks for a work in the style of da Vinci, then the ordinary
person test could be used. If the ordinary person points to an already existing work of da
Vinci that would match the result and satisfy the command given, then the user would be held
liable.
The contributory liability of the AI developer will be universal for the infringement caused
by the AI. The AI developer is the one who used such datasets, which had in them the original
work of creators, which was taken without the permission of the relevant owner and used to
train the AI, empowering the AI to give reproductions resembling the work and hence
infringing the exclusive rights of the author. The AI developer could have used only non-
copyrighted work or licensed the use of the copyrighted work from the owner. Then, there
would have been no scope of liability upon him. The principles of attribution of liability also
suggest that correspondence between risks, benefits, and responsibilities should guide the
liability69. Legal responsibilities should be allocated proportionately to the nature and scope
of risks each stakeholder( user and the AI developer) bears and the benefits they gain within
the generative AI ecosystem. The developer is the most suitable for bearing losses as he is the
one who charges for the services of the AI and, going by the vicarious liability principles, the
key person with the deep pockets to recover the damages of the victim. The Chinese court in
69
Yang Lu, ‘Artificial Intelligence: A Survey on Evolution, Models, Applications and Future Trends’ (2019) 6
Journal of Management Analytics 1.
Page | 19
Shanghai Character License Administrative Co., Ltd. (SCLA) v. tab (anonymity preserved)
held that there is a third-party due diligence on the service provider who uses AI on their site,
even though it did not create the generative AI itself. The Court’s findings that (i) similarity
between the output of a generative AI system and pre-existing works can amount to an
infringement of the rights in those pre-existing works, and (ii) the operator of a website
offering that an AI system can be liable for those outputs guide that due diligence has to
perform safeguarding the rights of others.
The paper proposes a system of dual liability with the AI developer as the primary liable
party and the user as a secondary liable party. After judging the user's intent, the secondary
party could be added to the proceedings based on the ordinary person test. The AI developer
will be liable under strict liability. There have been cases in which the AI developer develops
a contract with the user, granting him the entire ownership over the content produced by the
AI to escape liability. Still, it is imperative to understand that if no original work had been
used in the training of AI, no such result would have ever been produced.
CONCLUSION: REIMAGINING COPYRIGHT FOR THE AI ERA
The development of AI is an essential and integral step in the evolution of society. This trend
of technological advancements in the 21st century has led to vast changes in the food we eat
and how we think. The uses of AI as an autonomous problem-solving body have far-reaching
implications, like developing a new medicine that could treat cancer or creating a mechanism
to mitigate the results of global warming. This machine has a way of thinking inspired by the
human mind, containing neural networks empowering it to find possible solutions to
impossible questions. These uses of AI should be kept in mind while developing policies that
regulate the interaction between AI and copyright. Safeguarding an author's exclusive rights
should be the key subject, inviting innovative policies that balance protecting the rights and
the development of technology. This paper establishes that a parallel understanding of the
authorship of AI-generated content can help understand copyright infringement liability. After
discussing the position of various stakeholders, this paper points to a co-authorship between
the user and AI developer. While discussing the same, this paper puts forward the view that
AI being considered a legal person is now not far-fetched. The paper establishes the parallel
relation between authorship and liability by discussing the various liabilities introduced due
to the infringement. The paper contends that a joint liability can be introduced on the
Page | 20
principles of attribution of liability between the user and AI developer, considering the AI
developer as the primary and secondary liable party. The user being held liable shall be based
upon the ordinary person test to check the user's intention behind the prompt, which
generated an infringing result. The paper suggests the emergence of contractual licensing
between the association of the authors and the AI developers, granting remuneration to the
authors and the data to the developers. The paper proposes using preventive measures by the
developers, undermining specific prompts that have a significant chance of causing an
infringement. The paper encourages the courts to grant only injunctions and ask for technical
alterations due to the lawsuits, rather than granting hefty sums, discouraging the exploration
of AI. The paper finds it essential for the government to make changes to the copyright laws
to enable them to handle cases with AI as the subject matter, keeping in view that this is only
the early stages of AI and is yet to evolve. This paper encourages using metadata trails,
prompt logs, and watermarks to understand and establish liability for the infringing part. The
paper urges policymakers to make copyright management information compulsory against
each copyrighted work, making tracing liability flow across players easier. The upcoming era
of development is imperative, and the policy decisions taken will shape our future. Subjects
like copyright are crucial in developing a research-oriented mindset in society.
Page | 21