Reading time: 30 minutes

Written by Joel Ko Chin Kye | 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 Joel Ko, argues that it is important for laws to be developed to accommodate the interplay of generative AI and copyright law. In doing so, Joel studies whether copyright can subsist in works purely generated by generative AI be treated; and whether we should allow works derived from AI-generated content to be copyrightable, particular in situations where there is sufficient human editorial control.

Introduction

This report focuses on copyright ownership relating to artificial intelligence (“AI”). There are two main issues, namely (1) whether copyright can subsist in works purely generated (“AI-generated works”) by generative artificial intelligence (“generative AI”) be treated; and (2) potentially allowing works derived from AI-generated content, where there is sufficient human editorial control (“AI-assisted works”), to be copyrightable. 

In this paper, section II looks at the conceptual underpinnings for copyright law and artificial intelligence, and how there are overlapping considerations between these two areas. Section III focuses on the ownership of AI-generated works, which focuses on whether AI-generated works deserve copyright protection. Section IV analyses whether users can rely on the derivative works framework to assert copyright protection over works derived from AI-assisted works, before section V concludes this paper by consolidating the proposed recommendations to the current regulatory regime.

Generative AI has been experiencing rapid growth in the global landscape. Applications such as Sora, Midjourney and StableDiffusion have taken over the artistic landscape with their ability to generate images and videos.[1] In fact, this meteoric rise can be encapsulated as follows: “I had expected the technology to take another two to three years. As it happened, it took place faster than I imagined. Within 10 months, OpenAI had announced the Sora platform to produce AI-generated video.”[2] Generative AI has invigorated the artistic scene by enabling rapid ideation, accelerating the creative process.[3] This reduces production time and costs, leading to greater efficiency.[4]

However, one uncertainty throughout this meteoric rise of generative AI relates to intellectual property (“IP”) rights.[5] For copyright, the creative process of authorship is typically protected by copyright laws to preserve cultural values – to fuel society’s desire for artistic creativity.[6] Copyright seeks to protect the expression rather than the idea behind it.[7] This way, three objectives– rewarding and recognizing authors, incentivising the creation of new works and benefitting society through creative works – can be achieved.[8]

Today, the rapid development and untapped capabilities of generative AI creates uncertainty as to how generative AI may fit within the existing realm of copyright laws.[9] Further, the potential impact of generative AI on artistic expressions cannot be understated.[10] It is therefore imperative that laws are developed to accommodate the increasing interplay and risks of generative AI and copyright law.

Key concepts

This section will provide an introduction to the copyright and AI and their place in Singapore. There will first be an introduction to: (A) the general overarching considerations of Copyright and AI; as well as (B) the existing regulatory regime for Copyright and AI.

General overarching considerations of copyright and AI

Here, the overarching considerations of both (1) copyright and (2) AI will be explained, followed by explaining how a (3) balance between these two considerations can be found.

Overarching Considerations of Copyright

Utilitarian notions: Intellectual property rights is premised on utilitarian notions, to maximise social welfare.[11] Specifically, to achieve this in copyright, copyright protection is accorded to authorial works to incentivise authors to engage in creative works without the risk of their work being appropriated.[12] This enables the public to read, enjoy and appreciate creative works.[13] Further, generative AI models are limited by its training data,[14] which means that they would not be able to keep up with recent developments in the artistic sphere.[15] If humans’ creative abilities are not engaged, art would only be confined to what is known.[16] However, the beauty of art is that artists can find new ways to showcase the unknowns, which Generative AI would not be trained on. 

Societal considerations: In addition, the duration of copyright across the world is typically limited to the author’s lifespan with an additional number of years and to avoid perpetual copyright protection.[17] One argument against perpetual copyright protection is that discourse within society may not be able to function, instead being “swallowed up in private property, lawsuits and prosecutions”.[18] Perpetual copyright has the potential to limit the free market, which makes it less healthy for overall societal interests.[19]

Preventing free-riding: Lastly, copyright serves protect the “author’s investment in the production of the work against unfair competition, especially against a competitor’s free ride and his parasitical undercutting of the author’s expenses by unauthorised and cheaper copying”.[20] This explains why there is often the requirement of originality which requires authors to expend a “substantial amount of skill or labour”.[21]

AI in the creative sphere

In the creative space, AI can be used to generate creative works based off the existing data it is trained upon. It serves as a platform for authors to create works “more quickly and efficiently”.[22] In fact, its developing capabilities allow artists to “enhance their creative process, offering a fresh perspective and opening doors to uncharted artistic territories”.[23]

Striking a balance between copyright and AI

Historically, copyright law is meant to incentivise specifically human intellectual creation,[24] and hence the position of non-human creations (i.e. that of generative AI) is unclear at present. There are two opposing views regarding the protection of AI-generated material. While it has been argued that protecting AI-generated material may go against traditional notions of incentivising human beings to contribute new creative works, the counter-position is that protecting works generated by generative AI instead spurs investments into AI development and the commercialisation of AI output, which would contribute to economical advancements.[25] Ultimately, the government has to balance these considerations to ensure that it is both future-ready regarding the adoption of AI, yet ensure that the utilitarian notions of copyright are not eroded.[26]

Singapore’s current regulatory scheme for AI in respect of copyright issues

Law on copyright

Briefly, copyright law consists of several stages: (a) when copyright subsists; (b) when copyright is infringed; and (c) permitted uses of copyrighted material. This paper focuses on the first stage – that is, whether copyright can subsist in AI-generated material.

At present, since there is no specific legislation for AI works within the IP realm, it appears that AI-IP issues would fall within the existing provisions for IP law. The main instruments for copyright lie within the Copyright Act 2021 (“CA21”) and common law.[27] Copyright subsistence is governed by Parts 2 and 3 of CA21.[28] The reading of these provisions is often supplemented by common law, where Singapore courts often refer to IP decisions from various other jurisdictions,[29] given the trans-national nature of IP law. 

Existing AI regulatory regime

In Singapore, there is a balancing approach taken by the government towards AI regulation, between two key considerations. On one hand, they recognise the need to encourage innovation and deployment of societally-beneficial AI systems to protecting and promote human well-being[30] as well as generating economic gains and improving lives.[31] On the other hand, they have also recognised the need for increased clarity within copyright laws and generative AI.[32]

Issue 1A: Can AI users own AI-generated work?

The first challenge pertains to copyright protection of AI-generated work. Ordinarily, for a work to be original such that copyright can subsist,[33] a person must have had expended a substantial amount of skill or labour.[34] In Singapore, copyright ownership can only be given to natural persons.[35] This has implications on copyright subsistence – with no human author, there can be no “original work” capable of copyright protection.[36] If this is extended to AI-generated works, it suggests that there will be no copyright protection over AI-generated work given the lack of a human author.

Thus, the barrier in allowing copyright subsistence for AI-generated work arises from the difficulty in identifying the author for the AI-generated work. Since generative AI developers do not appear to claim copyright ownership over AI-generated outputs, [37] this paper evaluates this issue from the viewpoint of the AI user. Hence, this section focuses on these two areas: (1) whether purely AI-generated work can be protected by copyright; and (2) whether a users’ prompts into a generative AI tool can meet the originality requirements for copyright. 

Approaches in other jurisdictions

Whether copyright can subsist in purely AI-generated works

In the US, copyright cannot subsist in works that are autonomously generated by AI.[38] In Thaler v Perlmutter, it was held that since there is no human involvement for works autonomously generated by an AI tool, copyright cannot subsist in such work.[39] In Australia, the Australian High Court held that a phone directory which was merely an output of a computer programme without human intervention was not original and hence copyright cannot subsist within it.[40] Applying this principle to AI-generated work, autogenerated AI output itself cannot be protected by copyright in Australia. A similar position is taken in India – the Delhi High Court rejected a copyright claim on the basis that a natural person must have been the author, and not an artificial person.[41]

Whether an AI user’s prompts suffices for the originality requirement

In the US, a user’s prompts to a generative AI tool would not meet the threshold for copyright subsistence.[42] In Zarya of the Dawn, the US Copyright Office held that the Midjourney users are not authors for copyright purposes, even if they have added hundreds of prompts to reach their intended graphic, since they have no control over how Midjourney produces the content.[43] They referred to the decision of Burrow-Gills, where the author must be “one who has actually formed the picture”.[44] More recently, the Copyright Office stated that “when an AI technology receives solely a prompt from a human and produces complex … works in response, the ‘traditional elements of authorship’ are determined and executed by the technology – not the human user”.[45] For this reason, a user’s prompts would not allow them to take ownership over the AI-generated material. 

The position in the EU is less clear. The Court of Justice of the European Union (“CJEU“) held that copyright will only subsist if there is originality flowing from the “author’s own intellectual creation.”[46] The exact ambit of this is unclear, and it is likely up to individual EU member states to determine whether output of an AI-generative model can meet this requirement. 

In the UK, copyrighted work requires the author’s own intellectual creation.[47] Within the Copyright Designs and Patents Act 1988 (“CDPA”), the “person by whom the arrangements necessary for the creation of the work are undertaken” can possibly be the author of the work.[48] This alludes to copyright protection where a generative AI user inserts prompts into the application. It should be noted that there was a prior suggestion for Singapore to adopt this provision, either via subsidiary legislation or IPOS administrative guidelines, such that computer-generated works can also enjoy copyright protection.[49]

In Asia, it seems that copyright protection is accorded to AI-generated works on the basis of a users’ prompts. In China, while copyright subsistence requires original “intellectual achievement”,[50] the ambit of “intellectual achievement” appears to be a wide one. Following the Shenzhen Nanshan District Court in Shenzen Tencent v Shanghai Yingxun, AI-generated text could be protected by copyright since there were inputs via Tencent to arrange and select the data input, which was key to the output selection process.[51] In Beijing, the Beijing Internet Court held that a user’s input and prompt in StableDiffusion amounted to an intellectual contribution.[52] Similarly, within Asia, Japan appears to take the position that copyright ownership can be given to the user by evaluating the instructions and input prompts by the user.[53]

The current legal position in Singapore

As mentioned, under the current regime in Singapore, copyright can only subsist where there is human authorship.[54] Based on English common law, this requires a “causal link” between the computer-generated work and the human author.[55] However, there are several lingering uncertainties. First, whether purely AI-generated work can be protected by copyright, and second, what is the exact threshold for this causal link – specifically, when can a user assert that their prompts to the generative AI constitute a “substantial amount of skill or labour”.[56]

With this in mind, alongside foreign jurisprudence and the various theoretical considerations, it is posited that there should be enhancements to CA21 to address the two aforementioned uncertainties. Broadly, the recommendations are: (a) Output solely generated by AI should not be accorded copyright protection; and (b) AI-generated work cannot be protected on the basis of a user’s prompts. 

Recommendation 1: AI-generated works cannot be protected by copyright

Considerations: Copyright Protection of AI-generated material going against traditional notions of copyright law

Copyright is a property right which has moral connotations, meant to spur artistic creation and with utilitarian considerations for the greater good of society.[57] AI on the other hand, as an artificial person, would not require such incentives to create.[58] Hence, extending copyright protection over such works does not appear to fit with the purposes of intellectual property laws. 

It may be argued that extending copyright protection to AI-generated material would spur general AI innovation as AI developers would be able to monetize the outputs of their systems.[59] This increased output increases artistic creation and has the effect of promoting the greater good of society.[60] However, the effect is questionable since the AI’s code, which forms the basis of generative outputs, is already entitled to copyright protection.[61] This means that there is no further incentive for developers to innovate beyond what they already have. 

There are three further problems arising from extending copyright protection to AI-generated works. First, it may facilitate free-riding by AI developers. Second, AI-generated works are ultimately still limited by a finite level of training data. Third, there are also uncertainties relating to the duration of copyright for works from generative AI.

First, as the laws governing the use of generative AI training data is still in flux, the ambit of training data used is extremely wide,[62] encompassing a large range of media on web domains.[63] This essentially facilitates free-riding – where the AI developer “benefit[s] from someone else’s investment”[64] – relying on existing artistic efforts and enabling generative AI to create new works based on them.[65]

Second, while large amounts of training data are used for generative AI, the permutations from Generative AI are still finite. Generative AI, as a large language model, is trained on huge sets of data to generate an output which predicts what would best answer the prompt.[66] Given the limited training data for generative AI,[67] there is the possibility that the same user inputs could give rise to very similar if not identical output, especially for artistic works.[68] This has two problems: First, the AI user would essentially be free-riding on the AI, and second, this may give rise to copyright stockpiling – where the user simply asserts copyright protection over the multitude of expressions generated by varying prompts.[69] Thus, it is undesirable to allow users of generative AI tools to own copyright in the material auto-generated by AI.

Third, there are also problems relating to the duration of copyright. Further, this paper does not seek to suggest having a separate legal personality for AI for copyright subsistence. Based on existing laws, copyright protection is only for a finite period of time including the author’s lifespan and a set number of years.[70] If separate legal personality were to be adopted, the laws on duration of copyright would require a revamp. Thus, as there are still no attempts for horizontal regulation of AI,[71] this paper will only analyse copyright ownership from the point of view of the end-user.

Solution: Explicit rejection of copyright protection for AI-generated works

Drawing from the US position, copyright should not subsist in works that are solely generated by AI. As the US Copyright Office rightfully points out, since the human author does not have any control as to how the content is produced, they should not be credited for generating the expression by the AI tool. Thus, the position on AI-generated work where the author has no input should be incorporated into CA21 – specifically, via an express provision forbidding copyright protection for AI-generated works. Arguably, it is common for users to rely on a complex range of prompts to alter the AI-generated result into something bespoke. However, this should still not suffice for copyright protection as it would be difficult to ascertain a threshold for prompts that justify protection.

Recommendation 2: User prompts to generative AI will not meet originality requirement for protection of AI-generated material

Considerations: Copyright protection of AI-generated material would protect ideas and not just expressions

Originality requires a “substantial amount of skill or labour” to be expended, for copyright to subsist.[72] The rationale behind having such a requirement of substantial skill or labour is so as to prevent free-riding on a competitor’s effort and investment.[73] But there is a conundrum when the AI is the one which expended that substantial amount of skill or labour after the user’s input.

Further, copyright seeks to protect the expression rather than the idea behind it.[74] The US Copyright Office described prompts to generative AI as “instructions to a commissioned artist”.[75] If so, then it appears that “prompts” which the user is providing are mere ideas, which are taken by the “artist”, generative AI, to produce a form of expression that would ordinarily be accorded copyright protection. It would not cohere with the notions of copyright law if a user is given copyright protection over works on the basis of entering a prompt.

Solution: User prompts should not suffice for copyright ownership over AI-generated material

Thus, users’ prompts to the generative AI system should not suffice for copyright ownership over AI-generated material, similar to the US’ approach.[76] While UK and China possibly allude to giving copyright protection on to AI-generated works by referencing the user’s input to the AI,[77] it is posited that this should not be adopted. Such an approach essentially protects the prompts, which represent the author’s ideas, instead of the expression, which goes against copyright law.[78] Thus, by adopting the US approach, the existing Singapore position – where the expression and not the idea is protected – can be maintained.[79] Hence, the CA21 could be amended to make an express stipulation that human prompts to generative AI cannot give rise to copyright protection of the AI-generated work.

Issue 1B: AI-generated work as a starting point – Derivative works

While the earlier recommendations are meant to prevent copyright ownership for works purely generated by AI, there are still avenues for AI-assisted works to be copyrightable. One way to circumvent this is via relying on the “derivative works” framework, where a creator can create works based off existing work, and the alterations would be protected by copyright as long as the relevant requirements are met.[80]

Approaches in other jurisdictions: AI and derivative works

In Japan, based on the draft “Approach to AI and Copyright”, it was suggested that Article 2(1) of their Copyright Act allows protection of AI-related work, albeit requiring an evaluation of certain factors which includes, among others: human modifications to the AI-generated work, which allude to using generated AI work as derivative works.[81] In India, the Supreme Court held that for derivative works to get copyright protection, there must be substantial variation from the pre-existing work, and this variation cannot be trivial.[82] A mere copy of something in the public domain cannot support a copyright.[83]

In the US, for AI-assisted works, there must substantial edits to be made to the AI-generated work.[84] Derivative works is defined within the United States Copyright Act as a “work based upon one or more preexisting works”.[85] Copyright in such derivative works would only extend to “material contributed by the author of such work, as distinguished from the preexisting material employed in the work, and does not imply any exclusive right in the preexisting material”.[86] Under the US Copyright Office’s Copyright Registration Guidance, “AI-generated content that is more than de minimis should be explicitly excluded from the application”.[87] In other words, for AI-assisted works, authors must disclaim any substantially AI-generated content.[88] There must still be the requisite “degree of original, creative expression required to satisfy the test for originality in copyright”.[89] Excluding AI-generated content serves to limit the copyright claim.

The current legal position in Singapore

In the decision of Virtual Map v Suncool International, any original work which was derived from an infringement may nevertheless enjoy copyright.[90] This position on derivative works has been problematic, as it seems to incentivise infringement which goes against the general policy of copyright – to prevent unfair appropriation after an author has expended sufficient skill, labour and judgment in its creation.[91] However, where an AI-generated work is concerned, such concerns are militated against since the AI model is the one which produced the creation, and the owners of such AI systems do not themselves seek to assert ownership of AI-generated output.[92]

Recommendation 3: Copyright can subsist in derivative works of AI-generated works

It is suggested that users could claim copyright ownership of AI-assisted works, where AI-generated work is the starting point for users to build upon as a form of derivative works. For derivative works to be original, it has to fulfill several requirements, which include: (a) author expending sufficient skill, labour or judgment in creating the derivate work; and (b) there is a material alterationwhich makes the totality of the derivate work different from the source material. Lessons can be adopted from the US’ position, whereby AI-generated content has to be excluded from copyright registration, and only user content can receive copyright protection.[93]

The reason for denying copyright protection for AI-generated works, but allowing copyright protection for AI-assisted works is to prevent AI users from stockpiling copyright over the output of an AI.[94] Following the derivative works concept, AI users can only seek copyright protection over their contribution to the pre-existing AI work, instead of having full ownership of the AI-generated output.[95] This fulfills considerations of both AI and copyright law. First, AI development would continue being beneficial to society as they will continue providing society with new creations and ideas, without the risk of free-riding and abuse by human users. Second, this would also fulfill the utilitarian notion of copyright law – in that creators would still be incentivised to create new works, and in fact be incentivised to work together with generative AI, giving rise to a wider range of work for the public to enjoy and appreciate.[96]

Existing barriers to derivative works

However, while the derivative works approach is attractive in principle, there are two potential barriers for its adoption: (1) whether AI-generated works can qualify as published works which derivative works are typically based on; and (2) it may not be apparent as to what the AI-generated work relied upon was. An author could easily pass off AI-generated works as their own. Accordingly, there has to be caveats in place to allow copyright subsistence in derivative works based off AI.[97] These will be canvassed via two supplementary recommendations to ensure that the derivative works framework can be seamlessly adopted: (1) that AI-generated works are to enter into the public domain immediately; and (2) that watermarking of AI-generated works should be mandated.

Recommendation 3A: AI-Generated Works enter into public domain immediately
Considerations: Whether AI-generated works can qualify as existing works which derivative works are typically based on

For the derivative works framework to apply, there must be a prior existing work. As it stands, it is unclear if AI-generated works constitute prior existing works which derivative works can be based upon.[98] Though the difference appears minor, if AI-generated works are not considered existing works, then the user would essentially have copyright protection over the entire AI-generated work, together with the modifications made to the work.[99] However, if AI-generated works are considered existing works, then the user would only have copyright protection over the modifications made to the work, assuming the originality threshold is met. Therefore, the position of AI-generated works ought to be clarified.

Solution: Explicitly recognising AI-generated works as part of public domain

To ensure that the derivative work framework is applicable, it is posited that all AI-generated works should enter into the public domain once they are generated.[100] While this may seem counter-intuitive to how copyright is meant to protect artistic creations, it has been argued that since the true “author” of AI-generated work is the software, and the software has no legal rights to begin with, it would not go against any notions of copyright.[101]

With this, the end-user would be able to copyright any derivative works which are based off the AI-generated work as long as it satisfies the originality requirement.[102] Further, the utilitarian notion of copyright can be retained as human users would still be incentivised to engage in creative works based off AI-generated works.[103] Lastly, this also prevents free-riding on generative AI applications – users cannot simply claim copyright over AI-generated material which is put in the public domain.

At this juncture, it should be noted that for works to be considered in the “public domain”, it merely means that no exclusive intellectual property right exists in such works.[104] Thus, while there are concerns surrounding privacy of sensitive information which is inputted in AI systems,[105] there does not appear to be any extra privacy risk that users face just because AI-generated material is put in the public domain. In any case, users should continue exercising caution when inputting sensitive information into Generative AI models.[106]

Recommendation 3B: Watermarking of AI-generated content
Considerations: Ensuring clarity where AI-generated content is relied upon

Assuming the derivative works framework is adopted, then ordinarily, a user can assert copyright ownership by proving that there was sufficient skill, labour or judgment in creating the derivate work and that there is a material alteration which makes the totality of the derivate work different from the source material.[107] However, the next obstacle pertains to how the exact “source material” can be ascertained. Without any identifiers for AI-generated material, it would be difficult for the courts to make a value comparison as to the differences between the derivate work and the “source material”.[108] This system would be subject to abuse – users of generative AI may simply claim that the bulk of their derivate work was original seek to assert copyright over it, even when it was mostly AI-generated.[109] If the copyright system is abused in this manner, then creators may be disincentivized from continuing to engage in creative works.[110] This would go against the utilitarian notions of intellectual property law.[111]

Solution: Working with businesses to adopt watermarking capabilities

The proposed solution is mandating watermarking for AI-generated work. This would serve to: (1) identify content that came from generative AI which makes it easier to determine when a derivate work is copyrightable; and (2) it also prevents users from abusing Generative AI to readily claim ownership over AI-generated material. To implement these, steps taken in other jurisdictions may be of assistance.[112]

For the development of watermarks, there are already efforts by private corporations to develop digital watermarks.[113] That being said, the Singapore government can still play a part to assist in such development. Under the proposed Online Trust and Safety (“OTS”) Research Programme led by the Ministry of Communications and Information, one of the aims is to test watermark technologies albeit in the context of combating disinformation.[114] Nevertheless, such technology could potentially be extended to AI-generated work. With this initiative in place, the effects of new watermark technology could be evaluated before it is adopted for AI-generated work in Singapore. 

It is hoped that watermarking AI-generated content could potentially ensure that traditional artistes would continue exploring the boundaries of art and showcasing more artistic “unknowns”, which would maximise societal benefit.

Future Issue: User Removal of Watermarks

It should be noted that there are further problems pertaining to removal of watermarks. There are two possible solutions that can be explored in a future paper: First, for the Singapore government to continue collaborating with AI developers to develop digital watermarks that cannot be removed.[115] Second, to adopt a similar approach to the EU AI Act, which has an obligation for transparency and watermarking.[116] Potentially, penalties could be imposed on users who are found to have removed these watermarks for AI-generated material.

Amendments to existing regulatory framework

Therefore, while the “derivative works” framework is proposed to be adopted in Singapore, there are several other safeguards that need to be adopted. First, there should be explicit recognition within CA21 that AI-generated material would enter the public domain immediately after its generation. Second, there should be mandated standards for watermarking of AI-generated content, through governmental co-regulation with private stakeholders. Because watermarking is already a requirement alluded to under the EU AI Act, it would not add unreasonable compliance costs to developers, which may drive them out of operating within Singapore. Instead, this would ensure that copyright law can continue to fulfill its purposes even with the advent of AI.

Overall recommendation – Amendment to the CA21

This paper posits that for issues relating to copyright ownership for AI-generated material, the existing CA21 should be modified to reflect the changing nature of AI, especially concerning ownership of copyright. The proposed amendment under the CA21 would be: 

Section 110A: Copyright Subsistence in computer-generated works

Computer-generated works are considered published once produced.

Copyright cannot subsist in computer-generated works.

“computer-generated works” refers to work that is generated purely by computer programs, with no subsequent human editorial control over the output.

In subsection (2), “human editorial control” does not include prompts to generative computer programs.

Subsection (2) does not preclude prompts from being copyrightable if fulfills Sections 109 or 110.

With this amendment, the aforementioned situations can be covered. As a preliminary point, “computer program” is used here to keep in line with section 243 of CA21,[117] which is meant to cover AI programmes.[118]

First, where AI autonomously creates a piece of work, it is not copyrightable since there is no human control nor intellectual input into the output. Section 110A(2) covers this scenario, explicitly rejecting copyright subsistence in AI-generated material. Since the “true author” – the software – has no legal rights at present, there will not be any effect of disincentivising future creations.

Second, Section 110A(4) would prevent the user from asserting ownership of copyright by mere virtue of inputting prompts to the generative AI system. This solves the concerns of free-riding and monopolisation of “ideas” rather than “expressions”. 

Third, Section 110A(1) would go towards allowing the derivative works framework to apply for copyright subsistence in AI-assisted work. The allowance of AI-assisted works being protected would prevent free-riding of generative AI by artists, which has the effect of incentivising new works in the creative sphere.

Apart from amendments to CA21, the derivative works framework would also require watermarking of AI content. Since this is a nascent area, it is recommended that the government first take steps with private stakeholders towards developing robust watermarking. The government could build on existing efforts by OTS, to extend the development of watermarking from disinformation to all AI-generated material. For this area, co-regulation between the government and AI developers should continue to be prioritized.

Concluding remarks

The development of generative AI in the creative sphere has the potential to revolutionise AI regulation. If proper regulation is not taken, there could be impacts trickling down the creative scene. The conclusion drawn is that primary legislative reform is preferable to make it clear that autogenerated works by generative AI cannot be accorded copyright protection. That being said, copyright can still subsist in AI-assisted works. To facilitate this, a new amendment is suggested where all computer-generated work is published once produced. This would provide clarity to the courts regarding the applicability of the derivative works framework to AI-assisted works. Alongside this, the Singapore Government could work with private stakeholders to develop watermarking capabilities to better identify AI-generated content. With these amendments, it is hoped that both AI development and the fundamental thrusts of intellectual property can be preserved.

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] Robin Rombach, et al., “High-Resolution Image Synthesis with Latent Diffusion Models” 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), at pp 10674–10685. Cited by Eric Zhou, and Lee Dokyun, “Generative AI, Human Creativity, and Art” (2023) 3(3) PNAS Nexus at p 2.

[2] Singapore, Budget 2024 (February 26 2024, Tan Wu Meng).

[3] Ziv Epstein, Aaron Hertzmann, “Art and the science of generative AI” (2023) 380(6650) Science 1110-1111 at p 1111.

[4] Ziv Epstein, Aaron Hertzmann, “Art and the science of generative AI” (2023) 380(6650) Science 1110-1111 at p 1111.

[5] Infocomm Media Development Authority, Generative AI: Implications for Trust and Governance (Discussion Paper, 2023) at p 12.

[6] David Llewelyn and T Aplin, Cornish, Llewelyn & Aplin Intellectual Property: Patents, Copyright, Trade Marks and Allied Rights (Sweet & Maxwell, 9th ed, 2019) at para 10-042.

[7] Flamelite (S) Pte Ltd v Lam Heng Chung [2001] 3 SLR(R) 610 at [22]; Pioneers & Leaders (Publishers) Pte Ltd v Asia Pacific Publishing Pte Ltd [2010] 4 SLR 744 at [38]; David Llewelyn and T Aplin, Cornish, Llewelyn & Aplin Intellectual Property: Patents, Copyright, Trade Marks and Allied Rights (Sweet & Maxwell, 9th ed, 2019) at para 1-006. 

[8] Ashley Tham, “AI art and copyright: Does the law need to cover more?” Channel News Asia 16 March 2023 <https://www.channelnewsasia.com/singapore/ai-art-copyright-law-artificial-intelligence-authorship-originality-3339396> (accessed 8 April 2024).

[9] Gil Appel, Juliana Neelbauer, and David A. Schweidel, “Generative AI Has an Intellectual Property Problem” Harvard Business Review 7 April 2023 <https://hbr.org/2023/04/generative-ai-has-an-intellectual-property-problem> (accessed 8 April 2024).

[10] 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), p 6.

[11] Stephen R. Munzer, New Essays in the Legal and Political Theory of Property (Cambridge University Press, 2001). Cited in Ritu Paul, Intellectual Property Rights: A Utilitarian Perspective, available in https://ssrn.com/abstract=3842429 at pp 1-2.

[12] Ritu Paul, Intellectual Property Rights: A Utilitarian Perspective, available in https://ssrn.com/abstract=3842429 at p 18.

[13] Ritu Paul, Intellectual Property Rights: A Utilitarian Perspective, available in https://ssrn.com/abstract=3842429 at p 18.

[14] Marina Lammertyn, “50+ ChatGPT Statistics And Facts You Need to Know in 2024” Invgate, 8 February 2024 <https://blog.invgate.com/chatgpt-statistics#:~:text=ChatGPT%20limitations,-Despite%20ChatGPT%20improvements&text=ChatGPT’s%20knowledge%20is%20limited%20to,a%20cutoff%20of%20January%202022> (accessed 8 April 2024) at para 33.

[15] Avinash Saravanan, “The Limitations of Chat-GPT” Medium, 20 December 2022 <https://medium.com/@asarav/the-limitations-of-chat-gpt-8b73f5859bb4> (accessed 8 April 2024).

[16] See: Gillian M. Morriss-Kay, “The Evolution of Human Artistic Creativity” (2010) 216 J. Anat. 158-176, p 159.

[17] See, for example: Copyright Act 2021, s 28 (Singapore); Copyright, Designs and Patents Act 1988 c 48 (UK), section 12.

[18] Iain G. Mitchell QC “Back to the Future: Hinton v Donaldson, Wood and Meurose (Court of Session, Scotland, 28 July 1773)” (2009) International Free and Open Source Software Law Review, 1(2), 111 – 122, at p 118.

[19] Iain G. Mitchell QC “Back to the Future: Hinton v Donaldson, Wood and Meurose (Court of Session, Scotland, 28 July 1773)” (2009) International Free and Open Source Software Law Review, 1(2), 111 – 122, at p 120.

[20] Andreas Rahmatian, Copyright and creativity. The making of property rights in creative works (Edward Elgar, 2011) at pp 35 – 41. Cited from Andreas Rahmatian, “Originality in UK Copyright Law: The Old “Skill and Labour” Doctrine Under Pressure” (2013) IIC 44, 4–34, at p 5. See also: Amy B. Cohen, “When Does a Work Infringe the Derivative Works Right of a Copyright Owner?” (1999) 17 Cardozo Arts & Ent. L.J. 623 at p 643.

[21] AUVI Pte Ltd v. Seah Siew Tee and another [1991] 2 SLR(R) 786 at [32].

[22] Felipe Guimaraes e Equipe Aela, “Artificial Intelligence: How AI is Changing Art”, Aela, 1 April 2023 <https://aelaschool.com/en/art/artificial-intelligence-art-changes/> (accessed 8 April 2024).

[23] Greg Lucas, “Improve Digital Artist productivity with Generative AI”, Grand Canyon University, 8 Jan 2024 <https://www.gcu.edu/blog/performing-arts-digital-arts/improve-digital-artist-productivity-generative-ai#:~:text=Digital%20artists%20can%20use%20AI,doors%20to%20uncharted%20artistic%20territories> (accessed 8 April 2024). 

[24] Sam Ricketson, “Reflections on Authorship and the Meaning of a “Work” in Australian and Singapore Copyright Law” [2012] 24 SAcLJ at paras 33-35.

[25] 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 pp 9, 46. See also: Shlomit Yanisky-Ravid, “Generating Rembrandt: Artificial Intelligence, Copyright, and Accountability in the 3A Era – the Human-like Authors are already here – A new model” (2017) 659 Michigan State Law Review; Anders Guadamuz, “Do Androids Dream of Electric Copyright? Comparative Analysis of Originality in Artificial Intelligence Generated Works” [2017] IPQ 169; Robert Denicola “Ex Machina: Copyright Protection for Computer-Generated Works” (2016) 69 Rutgers University Law Review 251; Annemarie Bridy, “The Evolution of Authorship: Work Made by Code” (2016) 39 Colum J L & Arts 395. Cited in Simon Chesterman, Law and Technology in Singapore (Singapore Academy of Law, 2021) at para 16.013.

[26] See: Patrick Croskery, “Institutional Utilitarianism and Intellectual Property, (1993) 68(2) Chicago-Kent Law Review 631 at p 641.

[27] Copyright Act 2021.

[28] Copyright Act 2021, Parts 2 and 3.

[29] See: Global Yellow Pages Ltd v Promedia Directories Pte Ltd [2017] SGCA 28 at [17], referencing US decision; Flamelite (S) Pte Ltd and others v Lam Heng Chung and others [2001] 3 SLR(R) 610; [2001] SGCA 75 at [39], referencing Australian decision.

[30] Singapore Academy of Law, Law Reform Committee, Applying Ethical Principles for Artificial Intelligence in Regulatory Reform (Reform Paper, July 2020) at para 1.3 (Chairperson: Kannan Ramesh J and Charles Lim Aeng Cheng). See also: Ministry of Law, Why Copyright Matters to You, Ministry of Law <https://www.mlaw.gov.sg/why-copyright-matters/copyright/> (accessed 8 April 2024) & Intellectual Property of Singapore, Copyright Factsheet on Copyright Act 2021, IPOS, 24 November 2022 <https://www.ipos.gov.sg/docs/default-source/resources-library/copyright/copyright-act-factsheet.pdf>(accessed 8 April 2024).

[31] Smart Nation Singapore, Smart Nation and Digital Government Office, National Artificial Intelligence Strategy: Advancing our Smart Nation Journey (November 2019) <https://www.smartnation.gov.sg/files/publications/national-ai-strategy.pdf> (accessed 8 April 2024), at p 7.

[32] Infocomm Media Development Authority, Aicadium, Generative AI: Implications for Trust and Governance (Discussion Paper, 2023) <https://aiverifyfoundation.sg/downloads/Discussion_Paper.pdf> (accessed 8 April 2024) at p 21. See also: Cheryl Seah, An Evaluation of the Current Solutions to Address the Intellectual Property Challenges it Generates, Singapore Law Gazette (November 2023).

<https://lawgazette.com.sg/feature/generative-ai/> (accessed 8 April 2024).

[33] Copyright Act 2021, s 109(1)(a). 

[34] AUVI Pte Ltd v. Seah Siew Tee and another [1991] 2 SLR(R) 786 at [32], citing Macmillan & Co Ltd v Cooper (K&J) [1923] 40 TRL 186 at 190 (Lord Atkinson). 

[35] Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] SGCA 37 at [67].

[36] Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] SGCA 37 at [81]-[82].

[37] See: OpenAI, Terms of Use, OpenAI (31 January 2024). <https://openai.com/policies/terms-of-use> (accessed 8 April 2024); Midjourney, Terms of Service, Midjourney (7 March 2024) <https://docs.midjourney.com/docs/terms-of-service> (accessed 8 April 2024)

[38] Stephen Thaler v Shira Perlmutter (2023) United States District Court, District of Columbia, No. 22-CV-384-1564-BAH F.Supp.3d, at pp 5-7.

[39] Stephen Thaler v Shira Perlmutter (2023) United States District Court, District of Columbia, No. 22-CV-384-1564-BAH F.Supp.3d, at pp 5-7. See also: Gerard Kelly, US Court Rejects Copyright Application for “Creativity Machine” AI, Mason Hayes & Curran (14 Sep 2023) <https://www.mhc.ie/latest/insights/us-district-court-rejects-copyright-application-for-creativity-machine-ai> (accessed 8 April 2024)

[40] Telstra Corporation Ltd v Phone Directories Company Australia Pty Ltd [2015] FCAFC 156

[41] Navigators Logistics Ltd v Kashif Qureshi & Ors 254 (2018) DLT 307 at para 34. <https://indiankanoon.org/doc/102653512/> Cited in Pallavi Sondhi, India: A Brief Look At The Copyright Issues Raised By Generative AI, Mondaq, 5 July 2023 <https://www.mondaq.com/india/new-technology/1337528/a-brief-look-at-the-copyright-issues-raised-by-generative-ai> (accessed 8 April 2024).

[42] US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 37 CFR Part 202 (16 March 2023) <https://www.copyright.gov/ai/ai_policy_guidance.pdf> (accessed 8 April 2024) at p 4.

[43] Zarya of the Dawn (US Copyright Office: Registration # VAu001480196) <https://1-next-westlaw-com.libproxy.smu.edu.sg/Link/Document/Blob/If245d4102d2011eeac8a8868a0fa5d39.pdf?originationContext=document&transitionType=DocumentImage&uniqueId=0e7c879c-e11f-41d6-960d-03f584438823&ppcid=5fe54976f01a4fa08ef4c354e1904791&contextData=(sc.Search)> (accessed 8 April 2024) at p 8. Cited from Michael Landau, U.S. Copyright Office says that AI is not copyrightable— Zarya of the Dawn, in Alexander Lindey, Lindey on Entertainment, Publishing and the Arts (3rd ed, Clark Boardman Callaghan, 2023) at § 1:2.50.

[44] Zarya of the Dawn (US Copyright Office: Registration # VAu001480196) <https://1-next-westlaw-com.libproxy.smu.edu.sg/Link/Document/Blob/If245d4102d2011eeac8a8868a0fa5d39.pdf?originationContext=document&transitionType=DocumentImage&uniqueId=0e7c879c-e11f-41d6-960d-03f584438823&ppcid=5fe54976f01a4fa08ef4c354e1904791&contextData=(sc.Search)> (accessed 8 April 2024) at p 8-9. Cited from Michael Landau, U.S. Copyright Office says that AI is not copyrightable— Zarya of the Dawn, in Alexander Lindey, Lindey on Entertainment, Publishing and the Arts (3rd ed, Clark Boardman Callaghan, 2023) at § 1:2.50.

[45] US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 37 CFR Part 202 (16 March 2023) <https://www.copyright.gov/ai/ai_policy_guidance.pdf> (accessed 8 April 2024) at p 4.

[46] EU Copyright Directive (Fourth Chamber court), Infopaq International A/S v Danske Dagblades Forening (Case C-5/08) at para 35, 37. See also: Painer v. Standard Verlags GmbH, Axel Springer AG, Suddeutsche Zeitung GmbH, Spiegel-Verlag Rudolf Augstein GmbH & Co. KG, Verlag M. DuMont Schauberg Expedition der Kolnischen Zeitung GmbH & Co KG (C-145/10), [2012] ECDR 6. Cited from Andreas Rahmatian, “Originality in UK Copyright Law: The Old “Skill and Labour” Doctrine Under Pressure” (2013) IIC 44, 4–34, at p 8.

[47] Newspaper Licensing Agency Ltd. and others v. Meltwater Holding BV and others [2011] EWCA 890 Civ, [2012] RPC 1 at [20].

[48] Copyright, Designs and Patents Act 1988 c 48 (UK), section 9(3).

[49] Singapore Academy of Law, Law Reform Committee, Rethinking Database Rights and Data Ownership in an AI World (Reform Paper, July 2020) at para 2.76 (Chairperson: Kannan Ramesh J and Charles Lim Aeng Cheng). 

[50] Copyright Law of the People’s Republic of China (as amended up to the Decision of February 26, 2010, of the Standing Committee of the National People’s Congress on Amending the Copyright Law of the People’s Republic of China), Article 3; Beijing Internet Court Civil Judgment Li v Liu (2023) Jing 0491 Min Chu No. 11279 at p 12. Cited in Stanley Lai, et al., “Legal implications – Beijing Internet Court grants copyright protection to AI-generated artwork” National University of Singapore Centre for Technology, Robotics, Artificial Intelligence & the Law (March 2024) <https://law.nus.edu.sg/trail/legal-implications-beijing-internetcourt-copyright/#:~:text=A%20further%20notable%20quirk%20of,input%20of%20a%20natural%20person%E2%80%9D.> (accessed 8 April 2024).

[51] Shenzhen Nanshan District Court, Shenzen Tencent v Shanghai Yingxun [2019] Case No. Y0305MC No. 14010. Cited in Cooley, Copyright Ownership of Generative AI Outputs Varies Around the World, Cooley (29 January 2024) <https://www.cooley.com/news/insight/2024/2024-01-29-copyright-ownership-of-generative-ai-outputs-varies-around-the-world#:~:text=European%20Union%3A%20Ownership%20possible,relevant%20EU%2Dlevel%20case%20law> (accessed 8 April 2024).

[52] Cooley, Copyright Ownership of Generative AI Outputs Varies Around the World, Cooley (29 January 2024) <https://www.cooley.com/news/insight/2024/2024-01-29-copyright-ownership-of-generative-ai-outputs-varies-around-the-world#:~:text=European%20Union%3A%20Ownership%20possible,relevant%20EU%2Dlevel%20case%20law> (accessed 8 April 2024).

[53] Japan Copyright Act (Act No. 48 of May 6, 1970, as amended up to January, 1, 2022), Article 2(1); Squire Patton Boggs, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials” (12 March 2024) Lexology<https://www.lexology.com/library/detail.aspx?g=c2579f2b-b32b-46e5-b35d-6d0f54e03896#:~:text=On%20January%2023%2C%202024%2C%20the,comments%2C%20additional%20changes%20were%20made.> (accessed 8 April 2024).

[54] Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] 4 SLR 381 at paras 59-60.

[55] Jacob Turner, Robot Rules Regulating Artificial Intelligence (Palgrave Macmillan, 2019) at p 125. Cited from David Tan, “Generative AI and Authorship in Copyright Law”, National University of Singapore Centre for Technology, Robotics, Artificial Intelligence & the Law (September 2023) <https://law.nus.edu.sg/trail/generative_ai_and_authorship_in_copyright_law/#_ftnref12> (accessed 8 April 2024).

[56] See: AUVI Pte Ltd v. Seah Siew Tee and another [1991] 2 SLR(R) 786 at [32].

[57] Stephen R. Munzer, New Essays in the Legal and Political Theory of Property (Cambridge University Press, 2001). Cited in Ritu Paul, Intellectual Property Rights: A Utilitarian Perspective, available in https://ssrn.com/abstract=3842429 at pp 1-2.

[58] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) 165(5) University of Pennsylvania Law Review 1245 at p 1249, 1270; Daryl Lim, “AI & IP Innovation & Creativity in an Age of Accelerated Change” (2019) 52(3) Akron Law Review, 813-875, at p 840.

[59] Jens Martin Skibsted, “How IP laws can be reimagined to stimulate innovation” World Economic Forum<https://www.weforum.org/agenda/2024/02/how-ip-laws-can-be-reimagined-to-stimulate-innovation/> (accessed 8 April 2024). 

[60] Ministry of Culture, Community and Youth, “Arts contributing meaningfully towards a more cohesive society” MCCY <https://www.mccy.gov.sg/about-us/news-and-resources/speeches/2020/oct/arts-contributing-meaningfully-towards-a-more-cohesive-society> (accessed 13 January 2025).

[61] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) 165(5) University of Pennsylvania Law Review 1245 at p 1261

[62] Saffron Huang and Divya Siddharth, “Generative AI and the Digital Commons” The Collective Intelligence Project Working Paper<https://www.cip.org/research/generative-ai-digital-commons> (accessed 15 January 2025). Cited in Jack Hardinges, Elena Simperl and Nigel Shadbolt, “We Must Fix the Lack of Transparency Around the Data Used to Train Foundation Models” (2024) Harvard Data Science Review.

[63] Keivn Schaul, Chen Szu Yu, Nitasha Tiku, “Inside the secret list of websites that make AI like ChatGPT sound smart” (19 April 2023) The Washington Post <https://www.washingtonpost.com/technology/interactive/2023/ai-chatbot-learning/> (accessed 15 January 2025). Cited in Jack Hardinges, Elena Simperl and Nigel Shadbolt, “We Must Fix the Lack of Transparency Around the Data Used to Train Foundation Models” (2024) Harvard Data Science Review.

[64] Mark A. Lemley, “Property, Intellectual Property, and Free Riding” (2005) 83 Texas Law Review 1031 at p 1040.

[65] See: Andreas Rahmatian, Copyright and creativity. The making of property rights in creative works (Edward Elgar, 2011) at pp 35 – 41. Cited from Andreas Rahmatian, “Originality in UK Copyright Law: The Old “Skill and Labour” Doctrine Under Pressure” (2013) IIC 44, 4–34, at p 5. See also: Amy B. Cohen, “When Does a Work Infringe the Derivative Works Right of a Copyright Owner?” (1999) 17 Cardozo Arts & Ent. L.J. 623 at p 643.

[66] Harry Guinness, “How does ChatGPT work?” Zapier (6 September 2023) <https://zapier.com/blog/how-does-chatgpt-work/> (accessed 8 April 2024). 

[67] Ashwin Hariharan, “Generative AI vs. Traditional Machine Learning: What Sets Them Apart?” Medium (20 September 2023) <https://medium.com/egen/generative-ai-vs-traditional-machine-learning-what-sets-them-apart-419d093a0128#:~:text=However%2C%20as%20with%20any%20emerging,output%20generated%20by%20AI%20models.> (accessed 8 April 2024).  

[68] Charles Ross, “Does ChatGPT Give the Same Answer to Everyone?” Medium (20 March 2023) <https://medium.com/@charles-ross/does-chatgpt-give-the-same-answer-to-everyone-521e3e9355a4> (accessed 8 April 2024)

[69] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) 165(5) University of Pennsylvania Law Review 1245 at p 1261.

[70] Copyright Act 2021, s 28 (Singapore).

[71] See, for example: EU AI Act.

[72] See: AUVI Pte Ltd v. Seah Siew Tee and another [1991] 2 SLR(R) 786 at [32].

[73] Andreas Rahmatian, “Originality in UK Copyright Law: The Old “Skill and Labour” Doctrine Under Pressure” (2013) IIC 44, 4–34, at p 12.

[74] David Llewelyn and T Aplin, Cornish, Llewelyn & Aplin Intellectual Property: Patents, Copyright, Trade Marks and Allied Rights (Sweet & Maxwell, 9th ed, 2019) at para [1-006]. See also: David Tan, “Generative AI and Copyright Infringement” National University of Singapore Centre for Technology, Robotics, Artificial Intelligence & the Law (January 2024) <https://law.nus.edu.sg/trail/generative-ai-copyright-infringement/> (accessed 8 April 2024).

[75] US Copyright Office Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence Federal Register 88(51) <https://www.govinfo.gov/content/pkg/FR-2023-03-16/pdf/2023-05321.pdf> (accessed 8 April 2024). Cited in Trisha Ray, “Can AI be creative? Global copyright laws need an answer” (18 December 2023) <https://www.atlanticcouncil.org/blogs/new-atlanticist/can-ai-be-creative-global-copyright-laws-need-an-answer/> (accessed 8 April 2024). See also: US Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” 37 CFR Part 202 (16 March 2023) <https://www.copyright.gov/ai/ai_policy_guidance.pdf> (accessed 8 April 2024) at p 4.

[76] Zarya of the Dawn (US Copyright Office: Registration # VAu001480196) <https://1-next-westlaw-com.libproxy.smu.edu.sg/Link/Document/Blob/If245d4102d2011eeac8a8868a0fa5d39.pdf?originationContext=document&transitionType=DocumentImage&uniqueId=0e7c879c-e11f-41d6-960d-03f584438823&ppcid=5fe54976f01a4fa08ef4c354e1904791&contextData=(sc.Search)> (accessed 8 April 2024) at pp 8-9. Cited from Michael Landau, U.S. Copyright Office says that AI is not copyrightable— Zarya of the Dawn, in Alexander Lindey, Lindey on Entertainment, Publishing and the Arts (3rd ed, Clark Boardman Callaghan, 2023) at § 1:2.50; US Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” 37 CFR Part 202 (16 March 2023) <https://www.copyright.gov/ai/ai_policy_guidance.pdf>(accessed 8 April 2024) at p 4.

[77] Copyright, Designs and Patents Act 1988 c 48 (UK), section 9(3). Cited in Cooley, Copyright Ownership of Generative AI Outputs Varies Around the World, Cooley (29 January 2024) <https://www.cooley.com/news/insight/2024/2024-01-29-copyright-ownership-of-generative-ai-outputs-varies-around-the-world#:~:text=European%20Union%3A%20Ownership%20possible,relevant%20EU%2Dlevel%20case%20law> (accessed 8 April 2024).

[78] See: Global Yellow Pages Ltd v Promedia Directories Pte Ltd [2017] SGCA 28 at [15], [30].

[79] Consistent with Flamelite (S) Pte Ltd v Lam Heng Chung [2001] 3 SLR(R) 610 at [22]; Pioneers & Leaders (Publishers) Pte Ltd v Asia Pacific Publishing Pte Ltd [2010] 4 SLR 744; [2010] SGHC 211 at [38]

[80] Virtual Map (Singapore) Pte Ltd v Suncool International Pte Ltd [2005] 2 SLR(R) 157 at [12]-[14].

[81] Squire Patton Boggs, “Japan’s New Draft Guidelines on AI and Copyright: Is It Really OK to Train AI Using Pirated Materials” (12 March 2024) Lexology <https://www.lexology.com/library/detail.aspx?g=c2579f2b-b32b-46e5-b35d-6d0f54e03896#:~:text=On%20January%2023%2C%202024%2C%20the,comments%2C%20additional%20changes%20were%20made.> (accessed 8 April 2024).

[82] Supreme Court of India, Eastern Book Company v. D.B. Modak 2008 AIR SCW 49 at para 34. <https://indiankanoon.org/doc/1062099/> (accessed 8 April 2024). Cited in Pallavi Sondhi, “India: A Brief Look At The Copyright Issues Raised By Generative AI” (5 July 2023) Mondaq<https://www.mondaq.com/india/new-technology/1337528/a-brief-look-at-the-copyright-issues-raised-by-generative-ai> (accessed 8 April 2024)

[83] Supreme Court of India, Eastern Book Company v. D.B. Modak 2008 AIR SCW 49 at para 34 <https://indiankanoon.org/doc/1062099/> (accessed 8 April 2024).

[84] Zarya of the Dawn (US Copyright Office: Registration # VAu001480196) <https://1-next-westlaw-com.libproxy.smu.edu.sg/Link/Document/Blob/If245d4102d2011eeac8a8868a0fa5d39.pdf?originationContext=document&transitionType=DocumentImage&uniqueId=0e7c879c-e11f-41d6-960d-03f584438823&ppcid=5fe54976f01a4fa08ef4c354e1904791&contextData=(sc.Search)> (accessed 8 April 2024) at p 11-12. Cited from Michael Landau, U.S. Copyright Office says that AI is not copyrightable— Zarya of the Dawn, in Alexander Lindey, Lindey on Entertainment, Publishing and the Arts (3rd ed, Clark Boardman Callaghan, 2023) at § 1:2.50.

[85] United States Copyright Act 17 U.S.C., § 101.

[86] United States Copyright Act 17 U.S.C., § 103(b).

[87] US Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” (16 March 2023) U.S. Copyright Office, Library of Congress <https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence> (accessed 8 April 2024). See also: United States House of Representatives, “Oversight of the U.S. Copyright Office” Subcommittee on Courts, Intellectual Property, and the Internet (27 September 2023) <https://judiciary.house.gov/committee-activity/hearings/oversight-us-copyright-office> (accessed 8 April 2024).

[88] US Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” (16 March 2023) U.S. Copyright Office, Library of Congress <https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence> (accessed 8 April 2024).

[89] US Copyright Office, “Compendium of US Copyright Office Practices” US Copyright Office<https://www.copyright.gov/comp3/docs/glossary.pdf> (accessed 13 January 2025).

[90] Virtual Map (Singapore) Pte Ltd v Suncool International Pte Ltd [2005] SGHC 19 at [10].

[91] Stewart E. Sterk, “Rhetoric and Reality in Copyright Law” (1996) 94 Michigan Law Review 1197, at p 1207. Amy B. Cohen, “When Does a Work Infringe the Derivative Works Right of a Copyright Owner?” (1999) 17 Cardozo Arts & Ent. L.J. 623 at p 643-644.

[92] See: OpenAI, Terms of Use effective 31 January 2024, OpenAI (14 November 2023). <https://openai.com/policies/terms-of-use> (accessed 8 April 2024).

[93] US Copyright Office, “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence” (16 March 2023) U.S. Copyright Office, Library of Congress <https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence> (accessed 8 April 2024).

[94] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) 165(5) University of Pennsylvania Law Review 1245 at p 1261.

[95] See: Interlego AG v Tyco Industries Inc [1989] AC 217 at 262; Virtual Map (Singapore) Pte Ltd v Suncool International Pte Ltd [2005] 2 SLR(R) 157 at [13].

[96] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) University of Pennsylvania Law Review 165(5), 1245-1270, at p 1261.

[97] PropertyGuru Pte Ltd v 99 Pte Ltd [2018] SGHC 52 at [102]-[104].

[98] John M. Griem Jr, Judith Wallace, “Managing the Risk of Using AI-Generated Content in a World of Copyright Uncertainty: Are AI Content Generators and AI Generated Expressions “Derivatives” of Copyrighted Works?” Carter Ledyard 28 August 2023 https://www.clm.com/managing-the-risk-of-using-ai-generated-content-in-a-world-of-copyright-uncertainly-are-ai-content-generators-and-ai-generated-expressions-derivatives-of-copyrighted-works/ (accessed 8 April 2024).

[99] Isha Marathe, “Derivative or Not? IP Attorneys Weigh In on Major AI Art Copyright Class Action” Law 2 May 2023 <https://www.law.com/legaltechnews/2023/05/02/derivative-or-not-ip-attorneys-weigh-in-on-major-ai-art-copyright-class-action/?slreturn=20240313173501> (accessed 8 April 2024).

[100] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) University of Pennsylvania Law Review 165(5), 1245-1270, at pp 1265-1266.

[101] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) University of Pennsylvania Law Review 165(5), 1245-1270, at p 1265.

[102] John M. Griem Jr, Judith Wallace, “Managing the Risk of Using AI-Generated Content in a World of Copyright Uncertainty: Are AI Content Generators and AI Generated Expressions “Derivatives” of Copyrighted Works?” Carter Ledyard 28 August 2023 https://www.clm.com/managing-the-risk-of-using-ai-generated-content-in-a-world-of-copyright-uncertainly-are-ai-content-generators-and-ai-generated-expressions-derivatives-of-copyrighted-works/ (accessed 8 April 2024).

[103] Robert Yu, “The Machine Author: What Level of Copyright Protection is Appropriate for Fully Independent Computer-Generated Works” (2017) University of Pennsylvania Law Review 165(5), 1245-1270, at p 1266.

[104] Faith O. Majekolagbe, “Public Domain and Access to Knowledge” (2024) 31(1) Journal of Intellectual Property Law 1, at pp 6-7.

[105] Wu Xiaodong, Ran Duan, Ni Jianbing, “Unveiling security, privacy, and ethical concerns of ChatGPT” (2024) 2(2) Journal of Information and Intelligence 102 at p 109.

[106] See: Eugene Kim, “Microsoft warns employees not to share ‘sensitive data’ with ChatGPT” (1 February 2023) Business Insider<https://www.businessinsider.com/chatgpt-microsoft-warns-employees-not-to-share-sensitive-data-openai-2023-1> (accessed 17 January 2025). 

[107] PropertyGuru Pte Ltd v 99 Pte Ltd [2018] SGHC 52 at [102].

[108] Ron Karjian, “How to detect AI-generated content” (2 August 2023) TechTarget<https://www.techtarget.com/searchenterpriseai/feature/How-to-detect-AI-generated-content> (accessed 8 April 2024).

[109] Gagandeep Kaur, “Your employees are likely passing off AI-generated work as their own: Report” (16 November 2023) ComputerWorld<https://www.computerworld.com/article/1639738/your-employees-are-likely-passing-off-ai-generated-work-as-their-own-report.html> (accessed 8 April 2024); Alex Dan, “Watermarking the New Standard for Protecting Digital Art?” (30 August 2023) Linkedin<https://www.linkedin.com/pulse/ethicalai-ai-generated-watermarking-new-standard-protecting-dan> (accessed 8 April 2024). 

[110] See: Sam Ricketson, “Reflections on Authorship and the Meaning of a “Work” in Australian and Singapore Copyright Law” [2012] 24 SAcLJ at paras 33-35; Ashley Tham, “AI art and copyright: Does the law need to cover more?” Channel News Asia<https://www.channelnewsasia.com/singapore/ai-art-copyright-law-artificial-intelligence-authorship-originality-3339396> (16 March 2023).

[111] See: Stephen R. Munzer, New Essays in the Legal and Political Theory of Property (Cambridge University Press, 2001). Cited in Ritu Paul, Intellectual Property Rights: A Utilitarian Perspective, available in https://ssrn.com/abstract=3842429 at pp 1-2.

[112] Senator Sinema in the US Senate Committee Hearings. See: Peter Henderson, “Should the United States or the European Union Follow China’s Lead and Require Watermarks for Generative AI?” (24 May 2023); SFS Georgetown Journal of International Affairs<https://gjia.georgetown.edu/2023/05/24/should-the-united-states-or-the-european-union-follow-chinas-lead-and-require-watermarks-for-generative-ai/> (accessed 8 April 2024). 

See also: “US’ approach thus far has been one of guidance and steering, rather than embedded in law” (Derek du Preez, “President Biden seeks to legislate for ‘responsible AI development’ in the US” (24 July 2023) Diginomica <https://diginomica.com/president-biden-seeks-legislate-responsible-ai-development-us> (accessed 8 April 2024)).

[113] See, for example, Adobe: Emilia David, “Watermarking the future” (13 February 2024) The Verge < https://www.theverge.com/2024/2/13/24067991/watermark-generative-ai-deepfake-copyright> (accessed 8 April 2024).

[114] Lee Li Ying , “New $20m initiative in S’pore to develop tools to detect deepfakes and misinformation” (11 January 2024) Straits Times<https://www.straitstimes.com/singapore/politics/new-20m-initiative-in-s-pore-to-develop-tools-to-detect-deepfakes-and-misinformation> (accessed 8 April 2024).

[115] Lee Li Ying , “New $20m initiative in S’pore to develop tools to detect deepfakes and misinformation” (11 January 2024) Straits Times<https://www.straitstimes.com/singapore/politics/new-20m-initiative-in-s-pore-to-develop-tools-to-detect-deepfakes-and-misinformation> (accessed 8 April 2024).

[116] EU AI Act, Recital 70a. See: Bertin Martens, “The European Union AI Act: premature or precocious regulation?” (7 March 2024) Bruegel <https://www.bruegel.org/analysis/european-union-ai-act-premature-or-precocious-regulation#footnote14_9xwd2h3> (accessed 8 April 2024).

[117] Singapore Copyright Act 2021, s 243.

[118] Singapore Parliamentary Debates, Official Report (13 September 2021) vol 95 (Edwin C F Tong, Second Minister for Law) at para 55. Cited from 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 79.