Reading time: 29 minutes

Written by Kaelynn Kok Chu Shuen | Edited by Josh Lee Kok Thong

LawTech.Asia is proud to collaborate with the Singapore Management University Yong Pung How School of Law’s LAW4060 AI Law, Policy and Ethics class. This collaborative special series is a collection featuring selected essays from students of the class. For the class’ final assessment, students were asked to choose from a range of practice-focused topics, such as writing a law reform paper on an AI-related topic, analysing jurisdictional approaches to AI regulation, or discussing whether such a thing as “AI law” exists. The collaboration is aimed at encouraging law students to analyse issues using the analytical frames taught in class, and apply them in practical scenarios combining law and policy.

This piece, written by Kaelynn Kok, considers several legal issues around the use of copyrighted material in generative AI training. These include: (a) the appropriate balance Singapore should strike between protecting the rights of creators and supporting AI innovation; (b) whether Singapore’s existing copyright defences are applicable to protect AI developers from copyright infringement claims; and (c) the best approach for Singapore to take.

Introduction

In recent times, generative Artificial Intelligence (“AI”) has taken the world by storm, with technology companies aggressively investing in its development. Microsoft recently injected $10 billion into OpenAI, the creator of numerous well-known generative AI products like ChatGPT.[1] Amazon also concluded $4 billion investment in Anthropic, a US-based AI startup company, for the development of generative AI, earlier this year.[2] These figures underscore the growing potential of this sector and indicate its enduring presence for the foreseeable future.[3]

Generative AI presents a unique challenge in the ongoing clash between creators of copyrighted content and technology developers. AI systems are often trained on a wide variety of data from the Internet, inevitably ingesting copyrighted materials without consent from the rights-holder.[4] Creators and rights-holders are hence demanding recognition and compensation for the use of their work.[5]

This is not the first time our world has seen new technologies pose copyright-related issues—present-day conventional technologies, such as cable TV, video recording devices, and MP3 players, were once considered disruptive and faced significant legal scrutiny from the copyright sector.[6] However, what sets generative AI apart is the remarkable speed at which this technology has been launched and adopted.[7] In contrast, developments in law and policy tend to progress much more slowly.[8] Further, it is challenging to determine the appropriate balance among competing copyright interests during earlier stages of technological developments, especially given the potential for generative AI to compete with and significantly affect the livelihoods of creators.[9]

In light of these concerns, this report considers the following legal issues surrounding the use of copyrighted material in generative AI training:

  1. What is the appropriate balance for Singapore to strike between protecting the rights of creators and supporting AI innovation; 
  2. Whether Singapore’s existing copyright defences are applicable to protect AI developers from copyright infringement claims; and
  3. What is the best approach for Singapore to take, with reference to jurisdictions around the world. 

Understanding copyright challenges in generative AI training

The use of copyright materials in AI training

Generative AI refers to AI systems that generate content (text, images or other media) in response to a prompt.[10] To do so, they undergo machine learning which entails “learn[ing] the patterns and structure of their input training data and generat[ing] new data with similar characteristics”. [11] The most well-known generative AI systems are large language models (“LLM”) such as Chat GPT-4, Gemini, Claude and LLaMA.[12]

Generative AI uses machine learning, which “provid[es] systems the ability to automatically learn and improve on the basis of data or experience, without being explicitly programmed”.[13] This process involves the designing of an algorithm and feeding the AI system with vast datasets to analyse and identify patterns, correlations, and structures within the data.[14] The result is that the AI system is able to “make predictions on new and unseen data, devise solutions to a problem or generate content that mirrors the patterns and styles observed during its training”.[15] These capabilities are what makes generative AI so fascinating and novel—it is not confined to a predefined task and can exhibit creativity by creating new content and outputs from its training data.[16]

However, for generative AI to perform such tasks, access to data is crucial.[17] LLMs in particular require extensive training datasets in order to effectively respond to user queries and commands.[18] While AI developers often claim that LLMs are trained on publicly available material online, the inclusion of copyrighted materials—such as books, articles, artworks, and music—in these datasets, is inevitable.[19]  

For example, ChatGPT, a prominent example of generative AI today, is an LLM designed to process human inquiries or commands and generate comprehensive and insightful responses.[20] To achieve this level of interaction, it relies on access to an extensive collection of literary works, many of which are under copyright protection.[21] OpenAI, the developer of ChatGPT, has disclosed that early models like GPT-1 utilised sources such as BookCorpus which consists of over 7,000 unique unpublished books.[22] By the time GPT-3 was being trained, the datasets came from two Internet-based books corpora amounting to 357,000 titles.[23]

Although this topic remains subject to debate, one can see how the use of copyrighted works to train generative AI poses potential copyright infringement issues. The process of machine learning typically involves making copies of the work to train the AI model such as through web-scraping or digitisation of copyright works.[24] Such reproduction of copyrighted works risks infringing upon the reproduction rights of the copyright owner.[25]

This raises important questions that need to be addressed: First, assuming the use of copyrighted materials for AI training constitutes copyright infringement, should it be permissible under any copyright exceptions? Second, should the rights of creators, whose text and images train such models, be recognised, and compensated? 

Relevant policy considerations

In order to determine the best approach for Singapore in addressing these issues, it is imperative to understand the relevant policy considerations at play. At the core of intellectual property lies the principle of rewarding and incentivising human creativity.[26] Copyright laws prevent unauthorised copying or use of works to ensure that creators “receive due recognition and fair compensation for their intellectual creations”.[27] This promotes creativity and innovation, and positively impacts the economy.[28]

However, a delicate balance needs to be struck, wherein strong copyright protection must be accompanied by reasonable exceptions.[29] Such exceptions allow society to use copyrighted works for purposes that benefit society, such as education, cultural heritage preservation and innovation.[30]  

These considerations underscore the dilemma surrounding the use of copyrighted material in AI training. There lies a fundamental tension between: (1) protecting the rights of creators by ensuring that their original works are not used in AI training without recognition and compensation, which will encourage further creation and innovation; and (2) allowing AI developers to train AI systems on diverse datasets including copyrighted material, which will ensure higher quality AI systems and promote AI development.

Protecting the Rights of Creators 

Although the training datasets used for generative AIs remain undisclosed, a growing number of creators are noticing similarities between their own work and output produced by these systems.[31] This observation has led them to suspect that their works are being used in the training of generative AI systems without their knowledge or consent. Consequently, creators and right-holders are suing for copyright infringement and demanding compensation for such use of their works.[32] For instance, in January 2023, Getty Images commenced legal proceedings against Stability AI for copying over 12 million of its photographs to train the AI model without permission or compensation.[33] In July 2023, over 9,000 writers backed an open letter from the Authors Guild addressed to the heads of leading generative AI companies worldwide, such as Alphabet, OpenAI, Meta and Microsoft. The letter asked for these firms to “obtain consent, credit and fairly compensate writers for the use of copyrighted materials in training AI”.[34]

Aside from addressing the growing discontent amongst creators, there are compelling moral, social and political reasons to require consent and/or compensation to artists for use of their works in developing generative AI. First, generative AI systems are only able to mimic human literary and artistic expressions because they have the opportunity to analyse human works used as training data.[35] However, these systems are then able to create works that compete with human creators in literary and artistic fields.[36] It has hence been recognised that it is “only fair that human authors—who provide the source material for AI ingenuity—receive remuneration when AI productions finally kill the demand for the same human creativity that empowered the AI system to become a competitor in the first place”.[37]

Second, it is important to protect the rights of creators because unlike AI-generated productions, human-authored works have the ability to convey messages that delve into societal issues, offering critical insights into contemporary conditions.[38] While AI systems may replicate human creativity, they fall short in grasping the essence of human artworks and their societal relevance.[39] Supporting human authors thus fosters innovation and exploration in the literary and artistic realms.[40]

Third, from a socio-political perspective, the rapid growth of AI will inevitably lead to displacement of some human creators and cause considerable disruption in the literary and artistic market.[41] Compensation towards authors whose works are used in AI training will help provide financial support for these creators, facilitate their transitions into new roles, and fund artistic projects to encourage continuous human creativity.[42]

Ultimately, the promotion of human literary and artistic production is beneficial for the advancement of AI. Investing in human creativity ensures a continuous supply of diverse training material for generative AI systems, without which AI outputs risk stagnation and being stuck in a permanent loop.[43] Seen in this light, perhaps the interests of creators and AI developers are not diametrically opposing—supporting human creativity through mandating remuneration can align with the AI industry’s interests and promotes innovation in AI development.[44]

Promoting AI Development  

Although there are good reasons for protecting the rights of creators and ensuring consent and/or compensation for the use of their works to develop generative AI, it may not necessarily be desirable and beneficial to mandate this at the input stage. 

To operate effectively, AI systems need to have access to a diverse range of training materials which includes copyrighted material.[45] This was reflected in a submission made by OpenAI to the House of Lords, claiming that it would be impossible to create tools like ChatGPT without access to copyrighted material because “copyright today covers virtually every sort of human expression—including blogposts, photographs, forum posts, scraps of software code, and government documents”.[46] It added that limiting training material to out-of-copyright materials would produce inadequate AI systems.[47] Scholars have also echoed that contractual restrictions or a licensing-only model to regulate the use of copyrighted material to train AI systems will negatively impact the quality of the output.[48]

It is also impractical to obtain consent from every rights-holder as many cannot be identified or are unwilling to allow their works to be used as training data.[49] The result is that developers will be left with a limited and potentially biased selection of training materials. This limitation may exacerbate existing issues of bias and inaccuracy faced by generative AI systems.[50] Today, these challenges are already troubling for generative AI development. For instance, a 2023 study found that images generated with Stable Diffusion reinforce gender and racial stereotypes, potentially harming vulnerable communities if integrated into police sketch software.[51] AI Text generators like ChatGPT, Bing, and Bard have also been known to produce fabricated data, termed as “hallucinations”, which can mislead users.[52] The fear that restricting training datasets exclusively to publicly accessible, non-copyrighted materials may exacerbate these existing challenges are hence well-founded. 

Further, there have been calls for greater transparency in AI development worldwide, especially in the type of data used to train AI models.[53] These calls are mainly driven by concerns over the opaque nature of AI systems, often referred to as “black boxes”.[54] If the use of copyrighted material in AI training is deemed an infringement and compensation is necessary, the push for transparency would present a dilemma for developers: the more transparent they are about their training datasets, the greater the risk of facing copyright infringement claims. It is likely that developers will pivot towards a posture of increased secrecy concerning disclosure of their training data and the intricate mechanisms of their AI systems.[55]

Lastly, the imposition of licensing or contractual costs could pose significant barriers to entry for technology start-ups and small enterprises, hence limiting generative AI development to the very large and financially robust companies.[56] This could potentially lead to market dominance by these companies, stifling competition and innovation in the realm of AI. 

Therefore, while the arguments for mandating recognition and compensation to creators whose works serve as training data are compelling, enforcing this during the AI training phase has negative consequences that must not be overlooked. The proposed approach for Singapore to strike the right balance between protecting the rights of creators and promoting AI innovation is this: the law ought to have an exception that allows for all types of works, including copyrighted materials, to be used for AI training. At the same time, alternative approaches to ensure creators receive appropriate compensation should be explored. 

Copyright law exceptions for AI training: A look at Singapore and other jurisdictions

Singapore’s copyright law exceptions

Under Singapore’s Copyright Act 2021,[57] there is a general fair use exception[58] and exceptions for specific types of permitted uses—the most relevant for use of copyright-protected works for AI training is the computational data analysis exception.[59] This section will hence explore whether these exceptions afford adequate protection for the use of copyrighted materials in AI training, or whether amendments need to be made.

Computational Data Analysis Exception 

The Computational Data Analysis (“CDA”) exception was recently introduced to support Singapore’s Smart Nation’s objectives and grow its AI and technology sectors.[60] It has been said by Parliament that this exception can be relied on when training AI systems, without needing to seek permission from each right-holder.[61] However, a closer examination of the wording of the CDA exception would reveal that it may not be applicable for the use of copyright material in training AI systems.

CDA is defined non-exhaustively to include:[62]

  1. using a computer program to identify, extract and analyse information or data from the work or recording; and
  2. using the work or recording as an example of a type of information or data to improve the functioning of a computer program in relation to that type of information or data.

For the CDA exception to apply, one of the conditions that must be fulfilled is that the user must have “lawful access” to the copyrighted material.[63] Two examples given under the Act for what is “lawful access” is (a) “X does not have lawful access to the first copy if X accessed the first copy by circumventing paywalls”; and (b) “X does not have lawful access to the first copy if X accessed the first copy in breach of the terms of use of a database”.[64] However, the scraping of copyrighted materials from the Internet will more often than not be pirated, circumvent paywalls or violate the terms of use, hence constituting unlawful access.[65] Therefore, although the CDA exception is intended to apply for AI training, its practical application is constrained by this stringent prerequisite of lawful access.  

Fair Use Exception

The general fair use exception also does not provide adequate protection for the use of copyrighted material in machine learning.[66] Section 191 of Singapore’s Copyright Act sets out a non-exclusive list of four factors for a court to consider in determining whether an unauthorised use qualifies as fair and thus permitted:[67]

  1. the purpose and character of the use, including whether the use is of a commercial nature or is for non‑profit educational purposes;
  2. the nature of the work or performance;
  3. the amount and substantiality of the portion used in relation to the whole work or performance; and
  4. the effect of the use upon the potential market for, or value of, the work or performance. 

These four statutory factors are derived from the fair use doctrine in the US, hence the Singapore Court of Appeal stated that US jurisprudence on these factors “would be helpful in shaping our law”.[68]

The first and fourth factors, which have been the most influential factors in deciding whether the fair use exception applies,[69] weigh against a finding of fair use in the context of AI training. 

Under the first factor, courts have traditionally looked at whether the original and infringing works share the same purpose ie whether the use of the original works by the generative AI was transformative.[70] It can be argued that the use of copyrighted material for AI training, with the purpose of analysing and identifying patterns in data, has a transformative use and hence the first factor may weigh in favour of fair use.[71]

However, the US Supreme Court recently shifted its focus from transformative use to whether the copying was done for a commercial purpose.[72] This new focus on commercial motives has left the landscape uncertain as to whether the use of copyrighted material for AI training would qualify as fair use. If Singapore courts adopt this approach, it could negatively impact text-to-image generative AI models like DALL.E, Stable Diffusion, and Midjourney.[73] For example, if a user seeks an image for illustrative purposes and provides specific text prompts to a generative AI system for image production,[74] rather than obtaining a direct license from the original author, it may be held that it is for a commercial purpose such that the first factor weighs against a finding of fair use.[75]

Under the fourth factor, courts will likely consider the impact of the generative AI’s use of the copyrighted material on the market ie whether it threatens the livelihood of the original creator by competing with their works or the licensing market for their works.[76] If the copying had resulted in widespread revelations of significant portions of the original work, this would create a competing substitute, even if the purpose of copying was transformative.[77]  

Generative AI models today can reproduce a significant portion of an original work in response to a user’s text prompt. For instance, in the case of Authors Guild v Open AI Inc, allegations were made that ChatGPT could generate summaries of books and detailed outlines for purported sequels of books without authorisation.[78] This may hence be deemed as creating a competing substitute and a negative impact on the market.[79]

Therefore, given that our copyright exceptions do not sufficiently protect AI developers from copyright infringement during the AI training phase, there is a need to consider making amendments to the law. 

Approaches in other jurisdictions

We explore other jurisdictions’ approaches to understand their positions on this matter. As shall be seen, there is an absence of any international consensus on how to balance the interests of right-holders and support for AI innovation. Copyright defences vary across jurisdictions, ranging from broad allowances that permit AI developers to train generative AI systems with copyrighted materials to strict positions restricting such use of copyrighted material, while some jurisdictions await the larger jurisdictions to pave the way.[80]  

Japan’s “Nonenjoyment” Copyright Exception

Japan is a jurisdiction with an extremely broad copyright exception, earning it recognition as one of the world’s most AI-friendly countries and a “machine learning paradise”.[81] On 1 January 2019, Japan’s revised Copyright Act came into effect which saw the introduction of Article 30-4, a copyright exception titled “Exploitation without the Purpose of Enjoying the Thoughts or Sentiments Expressed in a Work”. Article 30-4 states that “[i]t is permissible to exploit a work, in any way and to the extent considered necessary, in any of the following cases, or in any other case in which it is not a person’s purpose to personally enjoy or cause another person to enjoy the thoughts or sentiments expressed in that work”.[82] The only qualification is that the use of copyrighted material must not “unreasonably prejudice the interests of the copyright owner in light of the nature or purpose of the work or the circumstances of its exploitation”.[83]

This Japanese copyright exception has been regarded as the “broadest TDM exception in the world” because (1) the exception applies to both commercial and non-commercial purposes; (2) it applies to any exploitation regardless of the right-holders reservations; (3) exploitation by any means is permitted; and (4) no lawful access is required.[84] Further, Japan’s Minister of Education, Culture, Sports, Science and Technology, Keiko Nagoaka indicated that AI companies in Japan can use “whatever they want” for AI training “regardless of whether it is for non-profit or commercial purposes, whether it is an act other than reproduction, or whether it is content obtained for illegal sites or otherwise”.[85]

However, following a recent clarification from the Japan Agency for Cultural Affairs, this provision may not be as broad as it seems. It allows broad rights to ingest and use copyrighted works for any type of information analysis, including for the training of AI models, without the need to seek consent from copyright holders,[86] because the government recognises the difficulty for AI developers to know whether the content its AI system has ingested is pirated or legitimate.[87] However, if the AI provider knew (or should have known) that it had ingested pirated/infringing materials, this increases the likelihood of their liability.[88] Further, if the AI provider knowingly/should have known that it ingested pirated materials, it should take steps to prohibit infringing copyright output, which could help defeat claims for contributory infringement.[89]

There has nevertheless been increasing dissatisfaction voiced by Japan-based content creators with the insufficient legal protection for their works.[90] Despite Japan’s push to enhance AI development through broad legal exceptions that allow the ingestion of copyrighted works, it struggles to ensure proper recognition and compensation for content creators.[91] This underscores the need for mechanisms that fairly compensate right-holders, even amidst the expansive copyright exceptions aimed at fostering AI innovation.

UK and EU Strict Copyright Exception

In the United Kingdom (“UK”) and the European Union (“EU”) have more restrictive copyright exceptions, reflecting a priority on providing robust protection for content creators.[92] For instance, the EU Directive on Copyright and Related Rights in the Digital Single Market of 2019 (the “Copyright DSM Directive”), permits text and data mining (“TDM”) for lawfully accessed works by any entity for any purpose,[93] subject to the TDM not being prohibited by restrictions imposed by rights-holders contractually or by technical means.[94] The allowance of contractual restrictions to TDM was put in place during the final stages of the Copyright DSM Directive’s adoption process as a concession to rights-holders.[95]

Stringent exceptions pose a problem because they limit the government’s flexibility to adapt and support innovation. For instance, in the case of the UK, which was the first jurisdiction to introduce a TDM exception in 2014, it allows TDM of lawfully accessed works solely for non-commercial scientific research.[96] However, with the growth in AI and the desire to support greater innovation in this area, the UK Intellectual Property Office sought to broaden this exception in 2021.[97] However, this faced resistance from creators who feared it would deprive artists of economic rewards for their works exploited by AI companies for commercial gain.[98] On 3 February 2023, the UK Minister for Science, Research, and Innovation announced that the proposal would not proceed.[99]

Given that Singapore’s CDA exception was only introduced in 2021 and there is currently no international consensus on this matter, it is still feasible and timely to consider amending the CDA exception.

US Fair Use Exception and Australia’s “Wait-and-see” Approach

As mentioned above, the US addresses AI model training through the fair use doctrine, lacking specific copyright exceptions regulating this issue. However, US courts have not ruled on the application of the fair use doctrine to generative AI training; many jurisdictions are awaiting the outcomes of several lawsuits involving creators and operators of these tools.[100]

One such jurisdiction is Australia. In Australia, there is neither an open-ended fair use exception nor specific exceptions for AI analysis or data mining.[101] AI developers in Australia can only rely on relatively limited fair dealing provisions, such as those for research and study, or narrow temporary reproduction exceptions initially introduced for online cache and RAM copies.[102]

Like many smaller jurisdictions, Australia has adopted a “wait-and-see” approach to AI and copyright.[103] It awaits US court decisions on ongoing legal actions against AI companies and monitors whether other countries enact legislative measures in this rapidly evolving technical landscape.[104] However, since Singapore has already implemented the CDA exception to provide protection for the use of copyrighted materials in machine learning, adopting a “wait-and-see” approach may not be as relevant anymore.

Recommendations: The way forward for Singapore

The Singapore government has recently introduced the Model AI Governance Framework for Generative AI, which identifies data as a critical component in AI model development which significantly impacts the quality of the model output.[105] It states that where the use of data for model training is potentially contentious, such as personal data and copyright material, it is “important to give business clarity, ensure fair treatment, and to do so in a pragmatic way”.[106]

On the copyright front, the framework suggests (1) creating an open dialogue with all stakeholders, (2) encouraging AI developers to “undertake data quality control measures” using “data analysis tools to facilitate data cleaning”, (3) more globally expanding the available pool of trusted data sets and (4) governments “working with their local communities to curate a repository of representative training data sets for their specific context (e.g. in low resource languages)”.[107]

However, these soft law measures are inadequate; we will likely still see mounting frustration among all parties involved. Legislative amendments will likely be necessary soon to address these issues effectively.

Presumption of lawfulness for AI developers unless the AI developer had knowledge of the system ingesting pirated materials

At present, Singapore has a CDA exception aimed at allowing AI developers to train their systems without infringing copyright. However, the CDA exception has the stringent requirement of “lawful access”[108] which is not realistic for an AI developer to comply with. As the Japanese government has recognised, the opaque nature of AI systems often leaves developers unaware of what the AI has ingested.[109] Further, the UK’s experience with overly strict data mining exceptions has shown that prolonged inflexibility leads to significant resistance and dissatisfaction from the creative community when attempting legal reforms.[110]

Furthermore, the fair use defence is unlikely to be applicable to AI system training.[111] Singapore’s decision to implement the CDA exception in 2021 also represents a divergence from US copyright law. Consequently, Australia’s wait-and-see approach for US copyright development is not as relevant to Singapore. 

In view of the foregoing, we are of the view that an amendment should be made to Singapore’s CDA exception. Drawing insights from Japan’s copyright exception, Section 244 should have an express provision which presumes lawful access on the part of AI developers. As such, there should be an additional provision, section 244A which states “X is presumed to have lawful access to the first copy unless X knowingly or should have known that the access was unlawful”. 

Implementing this change would provide AI developers with more flexibility to train systems without the fear of copyright infringement claims, thus fostering AI innovation. However, similar to the situation in Japan, there may be opposition and dissatisfaction from the creative community. Therefore, it is essential to establish a mechanism that adequately rewards content creators for their work.  

Consider the implementation of a statutory license or levy for generative AI systems

As mentioned above, the high transaction costs and risks of incomplete datasets hinder AI development renders solutions such as data-sharing agreements or royalty-based compensation models for content creators impractical.[112] It is also not inappropriate to impose this at the AI training stage.[113]

We hence recommend that Singapore can consider implementing a statutory license scheme or an AI levy for all developers and providers of generative AI systems. This would impose a general payment obligation for the use of copyrighted works of machine learning purposes, following which this remuneration paid to social and cultural funds of collective management organisations can be distributed to creators to foster and support human literary and artistic work.[114]

Although no jurisdiction has implemented such a mechanism, the concept of a statutory license however, is not entirely new to copyright law. For instance, particularly in the music sector, the US copyright system uses a “permitted-but-paid” regime aimed at limiting the exclusive rights of copyright owners for certain purposes.[115] This includes preventing monopolies in the music sector and reducing transaction costs associated with licensing sound recordings and television programs.[116]

In the EU, Article 5.2 b of the InfoSoc Directive introduced a remunerated private copying exception or limitation to compensate rightsholders for private reproduction.[117] This led to the establishment of levy systems, such as in Germany, where levies are imposed on devices allowing duplication and the funds are then distributed among creators.[118] In Italy, there is also a “permitted-but-paid” use model for engineering projects providing original solutions to technical problems.[119] These works are protected by a 20-year neighbouring right, allowing authors to be compensated for unauthorised for-profit use.[120] This aims to foster technical progress and prevent safer techniques from being monopolised by creators or rightsholders.[121]

Therefore, statutory licenses maximize the use of copyrighted content for machine learning while ensuring authors are compensated for commercial use of their work. However, given that no jurisdiction has adopted this yet, this limits Singapore’s ability to harmonise its laws with others. Singapore may hence wish to adopt a wait-and-see approach regarding the implementation of a statutory license. 

Concluding observations

Any assessment of the suitability of amendments must consider the particular features of the Singapore context: 

  1. First, the presence of a CDA exception shows that Singapore is positioned favourably to further encourage AI innovation. However there needs to be a less stringent requirement of “lawful access”.
  2. Second, with the increasing dissatisfaction among creators, Singapore may consider implementing a statutory licensing system for generative AI developers. Nevertheless, Singapore could opt for a “wait-and-see” stance on this, considering larger jurisdictions have not yet adopted similar measures on the copyright front.  

Editor’s note: This student’s paper was submitted for assessment in end-May 2024. Information within this article should therefore be considered as up-to-date until that time. The views within this paper belong solely to the student author, and should not be attributed in any way to LawTech.Asia.

[1] Dina Bass, “Microsoft Invests $10 Billion in ChatGPT Maker OpenAI”, Bloomberg <https://www.bloomberg.com/news/articles/2023-01-23/microsoft-makes-multibillion-dollar-investment-in-openai#xj4y7vzkg> (accessed 14 April 2024). 

[2] Amazon, “Amazon and Anthropic deepen their shared commitment to advancing generative AI”, Amazon<https://www.aboutamazon.com/news/company-news/amazon-anthropic-ai-investment> (accessed 14 April 2024). 

[3] Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797.

[4] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70; Daniel O’Leary, “Artificial Intelligence and Big Data”, (2013) 28 IEEE Intelligent Systems 96-99; Paul Zikopoulos, et al. Harness the Power of Big Data: The IBM Big Data Platform (McGraw Hill, 2012). Cited by Simon Chesterman, Good Models Borrow, Great Models Steal: Intellectual Property Rights and Generative AI (Law Working Paper No 2023/025) (National University of Singapore Law, 2023), at p 3.

[5] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 12; Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70.

[6] Pamela Samuelson, “Generative AI meets copyright” (2023) 381(6654) Science 158-161, at p 159.

[7] Pamela Samuelson, “Generative AI meets copyright” (2023) 381(6654) Science 158-161, at p 159.

[8] Pamela Samuelson, “Generative AI meets copyright” (2023) 381(6654) Science 158-161, at p 159.

[9] Pamela Samuelson, “Generative AI meets copyright” (2023) 381(6654) Science 158-161, at p 159.

[10] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 3.

[11] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 3. 

[12] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 3. 

[13] 15 U.S. Code § 9401. Cited by Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797. 

[14] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 69.

[15] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70. 

[16] Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797. 

[17] Huw Roberts, “The Chinese Approach to Artificial Intelligence: An Analysis of Policy, Ethics, and Regulation,” et al. (2021) 36 AI & Society, 59 – 77. Cited by Simon Chesterman, Good Models Borrow, Great Models Steal: Intellectual Property Rights and Generative AI (Law Working Paper No 2023/025) (National University of Singapore Law, 2023), at p 3.

[18] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70; David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, para 8.

[19] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70; Daniel O’Leary, “Artificial Intelligence and Big Data”, (2013) 28 IEEE Intelligent Systems 96-99; Paul Zikopoulos, et al. Harness the Power of Big Data: The IBM Big Data Platform (McGraw Hill, 2012). Cited by Simon Chesterman, Good Models Borrow, Great Models Steal: Intellectual Property Rights and Generative AI (Law Working Paper No 2023/025) (National University of Singapore Law, 2023), at p 3.

[20] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 8; Alec Radford, et al. “Language Models are Unsupervised Multitask Learners” (Technical Report) (OpenAI, 2019). <https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf> (accessed 14 April 2024). Cited by Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797.

[21] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 8; Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797. 

[22] Tremblay v OpenAI Inc, Case No. 3:23-cv-03223 (California District Court, 2023) at [28]-[35]. Cited by David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 8.

[23] Tremblay v OpenAI Inc, Case No. 3:23-cv-03223 (California District Court, 2023) at [28]-[35]. Cited by David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 8.

[24] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70; Pamela Samuelson, “Generative AI meets copyright” (2023) 381(6654) Science 158-161, at p 159. 

[25] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 11; Zeynep Ulkü Kahveci “Attribution problem of generative AI: a view from US copyright law”, (2023) 18(11) Journal of Intellectual Property Law & Practice, at p 797.

[26] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72. 

[27] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72.

[28] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72. 

[29] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72. 

[30] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72. 

[31] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 12. 

[32] David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 12; Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 70. 

[33] Getty Images (US) Inc v Stability AI Inc Case 1:23-cv-00135 (Delaware District Court, 2024). 

[34] Will Bedingfield, “The Generative AI Battle Has a Fundamental Flaw” Wired (25 July 2023) <https://www.wired.co.uk/artificial-intelligence-copyright-law?verso=true> (accessed 14 April 2024). Cited by David Tan, “Generative AI and Copyright Part 1: Copyright Infringement” [2023] SAL Prac 24, at para 14. 

[35] Marin Senftleben, Laurens Buijtelaar, “Robot Creativity: An Incentive-Based Neighboring Rights Approach”, available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3707741.

[36] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 69. 

[37] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1538. 

[38] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1538. 

[39] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1538. 

[40] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1538. 

[41] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1539.

[42] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1540; Joshua Gans, “Copyright policy options for generative artificial intelligence” VoxEU CEPR 4 April 2024 <https://cepr.org/voxeu/columns/copyright-policy-options-generative-artificial-intelligence> (accessed 14 April 2024)

[43] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1541; Reda Adel, “AI Innovation vs. Creator Rights: The Legalities of Training Models on Copyrighted Material” Medium 28 December 2023 https://medium.com/@adelalh777/ai-innovation-vs-creator-rights-the-legalities-of-training-models-on-copyrighted-material-678d5b97961c (accessed 14 April 2024).

[44] Tojin T. Eapen, et al., “How Generative AI Can Augment Human Creativity” Harvard Business Review August 2023 https://hbr.org/2023/07/how-generative-ai-can-augment-human-creativity> (accessed 14 April 2024); Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1541. 

[45] Nicola Lucchi, “ChatGPT: A Case Study on Copyright Challenges for Generative Artificial Intelligence Systems” (2023) European Journal of Risk Regulation, 1-23, at p 3. Anjana Susarla, “Generative AI could leave users holding the bag for copyright violations” The Conversation 22 March 2024 <https://theconversation.com/generative-ai-could-leave-users-holding-the-bag-for-copyright-violations-225760#:~:text=The%20legal%20argument%20advanced%20by,images%20like%20words%20and%20pixels> (accessed 14 April 2024).

[46] OpenAI—written evidence (LLM0113) House of Lords Communications and Digital Select Committee: Large language models (Committee Inquiry, 5 December 2023) at p 4. <https://committees.parliament.uk/writtenevidence/126981/pdf/> (accessed 14 April 2024).

[47] Nicola Lucchi, “ChatGPT: A Case Study on Copyright Challenges for Generative Artificial Intelligence Systems” (2023) European Journal of Risk Regulation, 1-23, at p 3; Sergio Brotons, “The Limitations of Generative AI, According to Generative AI” Lingaro<https://lingarogroup.com/blog/the-limitations-of-generative-ai-according-to-generative-ai> (accessed 14 April 2024).

[48] Sean M Fiil-flynn and others, “Legal Reform to Enhance Global Text and Data Mining Research” (2022) 378 Science 951. Cited by Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 71. 

[49] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 71. 

[50] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 71. 

[51] Nitasha Niku, Kevin Schaul, Szu Yu Chen, “These fake images reveal how AI amplifies our worst stereotypes” Washington Post 1 November 2023 <https://www.washingtonpost.com/technology/interactive/2023/ai-generated-images-bias-racism-sexism-stereotypes/> (accessed 14 April 2024).

[52] Roberto Mata v Avianca, Inc., Case no. 22-cv-1461 (San Diego District Court, 2023). See also: Cynthia A. Norton, Nancy B. Rapoport, “Doubling Down on Dumb: Lessons from Mata v. Avianca Inc” (2023) 42(8) American Bankruptcy Institute Journal 24-61, at p 24; Sai Anirudh Athaluri, et al., “Exploring the Boundaries of Reality: Investigating the Phenomenon of Artificial Intelligence Hallucination in Scientific Writing Through ChatGPT References” (2023) 15(4) Cureus. Cited in Robin Emsley, “ChatGPT: these are not hallucinations – they’re fabrications and falsifications” (2023) 9(52) Schizophrenia at p 1; Hussam Alkaissi, Samy I. McFarlane, “Artificial Hallucinations in ChatGPT: Implications in Scientific Writing” (2023) 15(2) Cureus at p 3.

[53] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 4. See, for example: United States, Generative AI Copyright Disclosure Bill. Cited in: Matt Milano, “A New Bill Aims to Bring Transparency to AI Training” WebProNews <https://www.webpronews.com/a-new-bill-aims-to-bring-transparency-to-ai-training/> (accessed 14 April 2024); United Kingdom Department for Science, Innovation & Technology, UK Office for Artificial Intelligence, “A Pro-innovation approach to AI regulation” UK Department for Science, Innovation & Technology, UK Office for Artificial Intelligence (Policy Paper, 3 August 2023) 

[54] Madalina Busuioc, Deirdre Curtin, Marco Almada, “Reclaiming transparency: contesting the logics of secrecy within the AI Act” (2023) 2 European Law Open 79-105, at p 82. See also: United Kingdom Department for Science, Innovation & Technology, UK Office for Artificial Intelligence, “A Pro-innovation approach to AI regulation” UK Department for Science, Innovation & Technology, UK Office for Artificial Intelligence (Policy Paper, 3 August 2023).

[55] Eliza Strickland, “Top AI Shops Fail Transparency Test Stanford transparency index rates Meta, OpenAI, and others on 100 indicators” IEEE Spectrum 22 October 2023 <https://spectru<m.ieee.org/ai-ethics> (accessed 14 April 2024).

[56] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 71. 

[57] Copyright Act 2021.

[58] Copyright Act 2021, section 191.

[59] Copyright Act 2021, section 244.

[60] Singapore Parliamentary Debates, Official Report (13 September 2021) vol 95 (Edwin C F Tong, Second Minister for Law) at para 58.

[61] Singapore Parliamentary Debates, Official Report (13 September 2021) vol 95 (Edwin C F Tong, Second Minister for Law) at para 55.

[62] Copyright Act 2021, section 243.

[63] Copyright Act 2021, section 244(2)(d); Singapore Parliamentary Debates, Official Report (13 September 2021) vol 95 (Edwin C F Tong, Second Minister for Law) at para 57.

[64] Copyright Act 2021, section 244(2)(d).

[65] David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 4; Chesterman, Good Models Borrow, Great Models Steal: Intellectual Property Rights and Generative AI (Law Working Paper No 2023/025) (National University of Singapore Law, 2023), at p 7.

[66] David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 5.

[67] Copyright Act 2021, section 191.

[68] Global Yellow Pages Ltd v Promedia Directories Pte Ltd and another matter [2017] 2 SLR 185, at [76].

[69] Barton Beeb, “An Empirical Study of U.S. Copyright Fair Use Opinions Updated, 1978-2019” (2020) 10 New York University Journal of Intellectual Property and Entertainment Law 1, at p 4. Cited by David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 16.

[70] Global Yellow Pages Ltd v Promedia Directories Pte Ltd and another matter [2017] 2 SLR 185, at [77]-[79]; Authors Guild v. Google, Inc., Case No. 13-4829 (2nd Circuit, 2015) at p 22.

[71] Donald Farmer, “Generative AI capabilities increase data analytics value” TechTarget 22 March 2024 < https://www.techtarget.com/searchbusinessanalytics/tip/Generative-AI-capabilities-increase-data-analytics-value#:~:text=Generative%20AI%20excels%20at%20identifying,to%20proactively%20develop%20mitigation%20strategies> (accessed 14 April 2024).  

[72] Andy Warhol v Goldsmith Case No. 21–869 (US Supreme Court, 2023) at p 2-3.

[73] McKinsey, “What is Generative AI?” McKinsey 2 April 2024 <https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-generative-ai> (accessed 14 April 2024).

[74] See, for example: Midjourney, “Quick Start” Midjourney <https://docs.midjourney.com/docs/quick-start> (accessed 14 April 2024).

[75] David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 14. 

[76] Global Yellow Pages Ltd v Promedia Directories Pte Ltd and another matter [2017] 2 SLR 185, at [84].

[77] Campbell v Acuff-Rose Music, Inc., Case No. 510 U.S. 569 (United States Court of Appeal, 6th Circuit, 1994) at p 571.

[78] Authors Guild v. Google, Inc., Case No. 13-4829 (2nd Circuit, 2015) at p 22. Cited by David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 18.

[79] David Tan, “Generative AI and Copyright Part 2: Computational Data Analysis Exception and Fair Use” [2023] SAL Prac 25, at para 15.

[80] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 72.

[81] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024); Artha Dermawan “Text and data mining exceptions in the development of generative AI models: What the EU member states could learn from the Japanese “nonenjoyment” purposes?” (2024) 27(1) Journal of World Intellectual Property 1-87 at p 53. 

[82] Japan Copyright Act (Act No. 48 of May 6, 1970) at Article 30-4.

[83] Japan Copyright Act (Act No. 48 of May 6, 1970) at Article 30-4.

[84] Artha Dermawan “Text and data mining exceptions in the development of generative AI models: What the EU member states could learn from the Japanese “nonenjoyment” purposes?” (2024) 27(1) Journal of World Intellectual Property 1-87 at p 46.

[85] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024); Ben Wodecki Jr., “Japan: Content Used to Train AI Has No IP Rights” AI Business 7 June 2023 https://aibusiness.com/data/japan-s-copyright-laws-do-not-protect-works-used-to-train-ai-> (accessed 14 April 2024); Ingrid Riehl “Japan Goes All In: Copyright Doesn’t Apply To AI Training” Business Information Industry Association 21 June 2023 https://www.biia.com/japan-goes-all-in-copyright-doesnt-apply-to-ai-training/> (accessed 14 April 2024). 

[86] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[87] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024). See also: James Vincent, “The scary truth about AI copyright is nobody knows what will happen next” The Verge (15 November 2022) <https://www.theverge.com/23444685/generative-ai-copyright-infringement-legal-fair-use-training-data>   (accessed 8 April 2024)

[88] K&S Partners, “Mitigating liability while copyright law catches up with Artificial Intelligence” Lexology 30 January 2024 <https://www.lexology.com/library/detail.aspx?g=77565328-e3f1-4b97-ab0f-990e861c3cf8> (accessed 14 April 2024); Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[89] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[90] Yomiuri Shinbun, “Japan Media Groups want Better Copyright Protection from AI” TheJapanNewshttps://japannews.yomiuri.co.jp/society/general-news/20230819-130476/> (accessed 14 April 2024); Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[91] Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[92] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 73.

[93] EU 2019 DSM Directive Art 3.

[94] European Union, Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC, Article 4.

[95] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 74. See also EU AI Act, Article 53(1)(c) and Article 53(1)(d).

[96] UK CDPA 1988, s29(A).

[97] United Kingdom Intellectual Property Office, Artificial Intelligence and Intellectual Property: Copyright and Patents: Government Response to Consultation (Consultation Outcome, 28 June 2022) <https://www.gov.uk/government/consultations/artificial-intelligence-and-ip-copyright-and-patents/outcome/artificial-intelligence-and-intellectual-property-copyright-and-patents-government-response-to-consultation> (accessed 14 April 2024). 

[98] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 74.

[99] Rachel Montagnon and Sungmin Cho, “UK Withdraws Plans for Broader Text and Data Mining (TDM) Copyright and Database Right Exception” Herbert Smith Freehills 1 March 2023 <https://hsfnotes.com/ip/2023/03/01/uk-withdraws-plans-for-broader-text-and-data-mining-tdm-copyright-and-database-right-exception/> (accessed 14 April 2024).

[100] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 76. 

[101] Intellectual Property Office of Singapore, When Code Creates: A Landscape Report on Issues at the Intersection of Artificial Intelligence and Intellectual Property Law (Report, 28 February 2024) at p 78.

[102] Rita Matulionyte, “Copyright and AI in Australia: 2023 in Review” Kluwer Copyright Blog 15 January 2024 <https://copyrightblog.kluweriplaw.com/2024/01/15/copyright-and-ai-in-australia-2023-in-review/> (accessed 14 April 2024).

[103] Rita Matulionyte, “Copyright and AI in Australia: 2023 in Review” Kluwer Copyright Blog 15 January 2024 <https://copyrightblog.kluweriplaw.com/2024/01/15/copyright-and-ai-in-australia-2023-in-review/> (accessed 14 April 2024); Australian Government Department of Industry, Science and Resources, Supporting Responsible AI: Discussion Paper (Discussion Paper, 1 June 2023). 

[104] Rita Matulionyte, “Copyright and AI in Australia: 2023 in Review” Kluwer Copyright Blog 15 January 2024 <https://copyrightblog.kluweriplaw.com/2024/01/15/copyright-and-ai-in-australia-2023-in-review/> (accessed 14 April 2024). 

[105] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 4.

[106] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 4. 

[107] AI Verify, Infocomm Media Development Authority, Proposed Model AI Governance Framework for Generative AI: Fostering a Trusted Ecosystem (Model Governance Framework, 16 January 2024) at p 9. Cited by Scott Warren, Joseph Grasser, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials?” Squire Patton Boggs Privacy World 12 March 2024 < https://www.privacyworld.blog/2024/03/japans-new-draft-guidelines-on-ai-and-copyright-is-it-really-ok-to-train-ai-using-pirated-materials/#:~:text=According%20to%20reports%2C%20in%20a,is%20an%20act%20other%20than> (accessed 14 April 2024).

[108] Copyright Act 2021, section 244(2)(d).

[109] See: Footnote 92. 

[110] See: para 46

[111] See: para 33 to 40. 

[112] Christophe Geiger, “Generative AI, digital Constitutionalism and Copyright: Towards a Stauttory Remuneration Right grounded in Fundamental Rights – Part 2” Kluwer Copyright Blog 19 October 2023 < https://copyrightblog.kluweriplaw.com/2023/10/19/generative-ai-digital-constitutionalism-and-copyright-towards-a-statutory-remuneration-right-grounded-in-fundamental-rights-part-2/> (accessed 14 April 2024).

[113] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1548. 

[114] Martin Senftleben, “Generative AI and Author Remuneration” (2023) 54 International Review of Intellectual Property and Competition Law 1535-1560, at p 1549.

[115] Jane Carol Ginsburg, “Fair Use for Free, or Permitted-but-Paid” (2014) 29 Berkeley Technology Law Journal 1446. Cited in Christophe Geiger, Vincenzo Iaia “The Forgotten Creator: Towards a Statutory Remuneration Right for Machine Learning of Generative AI” (2024) 52 Computer Law & Security Review at p 13.

[116] Sections 111, 112, 114, 115, 116, 118, 119, and 122 of the 1976 U.S. Copyright Act. Cited in Christophe Geiger, Vincenzo Iaia “The Forgotten Creator: Towards a Statutory Remuneration Right for Machine Learning of Generative AI” (2024) 52 Computer Law & Security Review at p 13.

[117] Directive 2001/29/EC of the European Parliament and of the Council of 22 May 2001 on the harmonisation of certain aspects of copyright and related rights in the information society, L 107, Article 5.2.

[118] Martin Kretschmer, Private Copying and Fair Compensation: An empirical study of copyright levies in Europe  (Research Paper No. 2011/9) (The Intellectual Property Office, 2011), at p 22.

[119] Christophe Geiger, Vincenzo Iaia “The Forgotten Creator: Towards a Statutory Remuneration Right for Machine Learning of Generative AI” (2024) 52 Computer Law & Security Review at p 13.

[120] Christophe Geiger, Vincenzo Iaia “The Forgotten Creator: Towards a Statutory Remuneration Right for Machine Learning of Generative AI” (2024) 52 Computer Law & Security Review at p 13.

[121] Christophe Geiger, Vincenzo Iaia “The Forgotten Creator: Towards a Statutory Remuneration Right for Machine Learning of Generative AI” (2024) 52 Computer Law & Security Review at p 13.