VidhiAagaz
International Journal ofLaw Management & Humanities
ISSN 2581-5369 · Open access
International Journal ofLaw Management & Humanities
Open accessArticlesVol. 9 No. VI (2026)pp. 1166–1184

The Dilemma of Authorship in the Age of Artificial Intelligence: Drawing the Line Between AI-Generated and AI-Assisted Works in Copyright Law from a Sri Lankan Perspective

MH
Menaka HarankahaSenior Lecturer Grade I at Department of Commercial Law, Faculty of Law, University of Colombo, Colombo, Sri Lanka
0views0downloads
Published 19 November 2026 · pp. 1166–1184 · Vol. 9 No. VI (2026)
Abstract

Generative artificial intelligence has destabilised an assumption embedded in copyright doctrine: that the person who communicates instructions to a machine is also the person who creates the resulting expression. This article examines where Sri Lankan copyright law should draw the line between AI-generated and AI-assisted literary and artistic works. Employing doctrinal and comparative methods, it analyses the Intellectual Property Act, No. 36 of 2003, leading Sri Lankan originality decisions and approaches adopted in the United States, the United Kingdom, the European Union and India. It argues that the statutory requirement of a physical-person author prevents attribution of authorship to an autonomous system, but does not preclude protection for sufficiently original human contributions embodied in an AI-mediated work. It proposes a five-stage Human Creative Contribution Test, proceeding through attribution, expressive specification, effective control and causation, originality, and protection and allocation. The proposed model avoids both automatic copyright for every prompt and indiscriminate denial of protection whenever AI has assisted a human creator. It further recommends evidential disclosure, clarification of partial protection and carefully framed statutory guidance rather than the immediate introduction of machine authorship.

Keywords
Artificial IntelligenceCopyrightAuthorshipOriginalityAI-Assisted WorksAI-Generated WorksSri LankaIntellectual Property
Contents
Full text

Introduction: the new authorship problem

Copyright has traditionally understood creation through a human relationship with expression. A novelist chooses the words of a paragraph; a photographer selects lighting, perspective and the moment of capture; a composer chooses the pattern of notes.1 Digital tools have long assisted those acts without becoming authors themselves.2 Generative AI complicates this description because a user can articulate an idea in ordinary language and receive a detailed picture, song, programme or article whose expressive particulars were not consciously designed by that user.3 The legal question is no longer simply whether a computer was used.4 It is which actor, if any, made the legally relevant creative choices realised in the particular output.5

The distinction matters commercially and constitutionally. Copyright gives exclusivity against others who reproduce protected expression, but an over-expansive claim can withdraw commonly generated material from public use.6 A categorical refusal to recognise any human creativity in AI-assisted projects would, conversely, penalise legitimate writers, designers, scholars and film-makers who employ new tools as part of their craft.7 The problem is acute where a final product contains both generated passages and human revisions, where repeated prompts yield unexpected alternatives, or where a person arranges numerous outputs into an authored composition.8 A defensible rule must attach protection to particular human contributions rather than treating an entire digital file as uniformly protected or unprotected.

Sri Lanka is well placed to frame the issue doctrinally because its copyright statute supplies two strong anchors. Section 5 defines an author as the physical person who created the work, while section 6 protects original intellectual creations.9 Yet neither provision expressly identifies the degree of influence necessary when probabilistic software generates the details.10 This article asks whether prompting, curation, modification and combination can amount to authorship; whether a developer, deployer, employer or commissioner can claim rights; and what a Sri Lankan court should require in evidence.11 Its principal claim is that authorship must be assessed at the level of expressive contribution, not at the level of mere technological participation.

The method is primarily doctrinal. The article reads the relevant Sri Lankan statutory provisions together, considers the domestic case law on originality, expression and derivative works, and then uses foreign statutes, judgments, copyright-office decisions and scholarship as persuasive comparative material.12 Comparative jurisdictions do not speak with one voice: the United Kingdom and India contain express computer-generated-work attribution rules, whereas United States administrative and judicial approaches insist on human authorship, and European originality doctrine asks whether expressive choices are free and creative.13 Rather than transplanting one model wholesale, the article identifies which elements can operate within Sri Lanka’s existing text.14

Conceptual distinctions

An AI-generated work, as the term is used here, is an output whose relevant expressive features were determined by a generative system without sufficient creative human determination of those features. An AI-assisted work is an output containing identifiable human-authored expression that AI has facilitated, transformed or helped implement.15 These are functional categories rather than descriptions of particular software. A single illustrated book may contain human-authored narrative, unprotected generated illustrations and a protectable original selection or arrangement.16 Likewise, a poem may contain independently authored lines combined with machine-suggested alternatives.17 The same workflow can therefore cross the boundary several times.

It is also essential to separate copyrightability from authorship, ownership and infringement. Copyrightability asks whether a qualifying original work exists; authorship identifies the natural person who created its protected expression; ownership allocates economic rights once such a work exists; and infringement examines unauthorised use of another protected work.18 A company may own rights without being the human author, and a purported AI output may infringe a pre-existing work even if that output itself receives no copyright.19 Confusing these propositions encourages the incorrect assertion that paying for a subscription, owning a model or commissioning a result automatically establishes authorship of all generated content.20

The statutory architecture of Sri Lankan copyright

The Intellectual Property Act, No. 36 of 2003, does not leave the identity of an author entirely open.21 Its definition of “author” expressly refers to the “physical person who has created the work”. This language is more decisive than an abstract reference to originality alone. A generative system is neither a physical person nor capable, under existing legislation, of holding the author-specific legal status contemplated by the Act.22 A corporation is likewise not the physical author merely because it supplied computers, funds or instructions; it may, in defined circumstances, be the owner of economic rights.23 The distinction is a statutory one, not a claim that machine outputs cannot be aesthetically valuable.

Section 6(1) protects literary, artistic and scientific works that are “original intellectual creations”, including books, articles, computer programs, music, audiovisual works, drawings and photographs. Section 6(2) grants protection by the sole fact of creation, irrespective of form, quality or purpose.24 Consequently, an AI-assisted work need not satisfy a special registration procedure or reach an elevated aesthetic standard.25 The relevant issue is whether its claimed expressive elements originated in sufficiently original human intellectual activity.26 A computer-assisted drawing and a conventionally drawn picture should not be subject to different standards merely because their tools differ.

Section 7 extends protection to translations, adaptations and other transformations, and to collections that are original through selection, coordination or arrangement. This provision is particularly significant for AI-assisted projects.27 A human translator who materially revises a machine translation may contribute copyrightable expression; a curator who constructs an imaginative narrative sequence from generated components may author that sequence.28 However, derivative-work protection is without prejudice to rights in pre-existing works.29 It does not confer ownership over embedded material simply because the compiler has arranged it creatively, nor does it validate unauthorised copying carried out through an AI service.

Section 8 excludes ideas, procedures, systems, concepts, principles, discoveries and mere data. A prompt that states an unprotectable idea or a general style cannot turn that idea or style into the user’s exclusive property.30 Conversely, the exact wording of an unusually original and sufficiently substantial human-written prompt may itself be a literary work, without giving its author copyright over an unpredictable image produced in response.31 The idea–expression distinction therefore operates twice: first in deciding whether the prompt is protected as text, and second in identifying who authored the resulting image or other content.32

Sections 9 and 10 distinguish economic from moral rights. The economic rights include reproduction, adaptation and communication; the moral rights concern attribution and prejudice to an author’s honour or reputation.33 Section 13 normally connects duration with the human author’s life, and section 14 distinguishes initial authorship from employment, commissioned and collective-work ownership rules.34 Section 15 creates a rebuttable presumption from an indicated human author’s name, but that presumption does not turn a misleading attribution into creative fact. Section 16 regulates written assignments and licences.35 These interconnected rules reinforce the practical importance of identifying the relevant human creator before deciding who may commercially exploit the work.

Originality and the contribution of Sri Lankan jurisprudence

Sri Lankan courts have already supplied a doctrinal bridge between classical originality and AI-assisted creation. In Wijesinghe Mahanamahewa v. Austin Canter, the Court of Appeal treated originality as concerned with expression that originates with an author rather than novelty in its absolute form.36 The Court of Appeal’s discussion in Vasantha Obeysekera v. Anthony Christopher Alles similarly associates original expression with the creative mind.37 Fernando v. Gamlath, drawing on University of London Press, emphasises that copyright protects the expression of thought rather than the novelty of ideas.38 None is an AI case, but together they provide criteria capable of being applied to modern creative workflows.

The crucial inference is that effort and skill do not themselves settle the authorship of every output. A person can spend many hours designing prompts, maintaining a model or clicking through alternatives without selecting the exact expressive features eventually produced by an algorithm.39 Such labour may matter economically and may itself generate separately copyrightable text, code or arrangements.40 But courts must still ask whether the claimed expression originated from that person’s intellectually creative decisions. This is consistent with the distinction drawn in Feist between originality and labour alone, and with the Canadian Supreme Court’s emphasis in CCH on skill and judgment rather than mechanical exercise.41

The benchbook Intellectual Property Adjudication in Sri Lanka, published by WIPO and the United Nations Development Programme, expressly recognises the statutory human-authorship requirement and observes that the question of AI-assisted output has not been resolved by Sri Lankan courts.42 It also interprets originality as involving independent creativity rather than mechanical copying.43 The benchbook further suggests that, because the Act makes no specific provision for computer-generated works, the author of such a work is the person who made the arrangements necessary for its creation.44 That observation should not, however, be transformed into a settled judicial holding that the person arranging a computer process must always be the author: section 5 requires actual creation by a physical person, and Sri Lankan legislation lacks the express computer-generated attribution deeming rule enacted in the United Kingdom.45 Attribution and the threshold of creative involvement therefore remain questions of interpretation and possible reform.46

A court should distinguish between protected expression that a person actually supplied, expressive details selected from a range of possibilities, and elements whose form was decided entirely by the system. It should also resist a crude quantitative standard: a single distinctive change can sometimes transform the expressive character of a work, while hundreds of routine alterations may not.47 The analysis is qualitative, contextual and directed to the particular features for which exclusivity is claimed.48 This approach also preserves the traditional low threshold of originality without collapsing it into mere causal involvement.49

Comparative approach I: the United States

American law currently offers the most elaborated administrative account of generative AI copyrightability. The Copyright Office’s March 2023 guidance requires applicants to disclose more-than-de-minimis AI-generated material in registration claims and to identify the human-authored portions.50 Its January 2025 report maintains that using AI in the creative process does not categorically defeat copyright, but that prompts alone, at the levels of control then considered, generally do not establish authorship of generated expressive details.51 Human selection, coordination, arrangement and meaningful modification can independently qualify.52 The governing proposition is contribution-specific copyright, not a blanket ban on AI-mediated works.

In Zarya of the Dawn, the Copyright Office differentiated the author’s text and arrangement of a comic from individual images produced through Midjourney.53 In Théâtre D’opéra Spatial, the Office’s Review Board upheld the refusal to register a generated visual work where the applicant would not disclaim substantial AI-produced material, notwithstanding extensive iterative prompting and assertions of subsequent editing.54 These administrative outcomes are fact-sensitive and are not judgments binding Sri Lankan courts.55 Their value lies in demonstrating that claiming the copyright of an overall compilation is different from claiming the copyright of each component image.

In Thaler v. Perlmutter, the United States Court of Appeals for the District of Columbia Circuit affirmed the refusal to register a work identified as autonomously created by an AI system, because human authorship was required. The court addressed the applicant’s stipulated claim of machine-only creation, not every conceivable collaborative human–AI process.56 That limit matters.57 It would be misleading to cite Thaler as proof that all outputs touched by generative AI are excluded from protection. The older decision in Burrow-Giles, recognising the photographer’s original creative contribution to a technologically produced photograph, illustrates that copyright can attach where a machine executes an author’s creative decisions.58

The American position is instructive but not conclusive for Sri Lanka.59 Both systems are rooted in human authorship, yet the precise statutory language, registration arrangements, moral rights and ownership rules differ. The Sri Lankan Act gives automatic protection upon creation and expressly defines the author as a physical person; it does not require American registration formalities.60 A Sri Lankan court should therefore borrow the substantive inquiry into human control and separable expression without importing a foreign administrative procedure as a condition of local subsistence.61

Comparative approach II: the United Kingdom and India

The United Kingdom takes a strikingly different legislative route. Section 9(3) of the Copyright, Designs and Patents Act 1988 deems the author of a computer-generated literary, dramatic, musical or artistic work to be the person by whom the arrangements necessary for its creation are undertaken.62 The legislation defines computer-generated works by reference to circumstances in which there is no human author.63 The provision was enacted before today’s generative models and invites difficult questions: are the necessary arrangements made by the programmer, deployer, prompt writer, model provider, or some combination?64 Causal responsibility is not necessarily the same as personal creative authorship.65

Nova Productions v. Mazooma Games offers a narrower warning against assuming that every user interaction constitutes authorship.66 The English Court of Appeal did not treat a player’s actions in operating a video game as authorship of the resulting frame images; the relevant creative arrangements were associated with the development of the game. Modern prompt-based systems are technically different, and the case does not allocate copyright in large language model outputs.67 Nevertheless, it reveals an enduring problem for any arrangement-based rule: the individual who triggers a result is not necessarily the person whose actions embody the creative conception of its expression.68

India’s Copyright Act, 1957, section 2(d)(vi), similarly identifies the author of a computer-generated literary, dramatic, musical or artistic work as the person who causes it to be created.69 That language appears hospitable to users, but causation can be distributed across model training, design, prompt formulation and deployment. Its existence does not establish that every AI-generated work in India automatically receives enforceable protection, nor does it resolve originality or competing claims.70 In contrast, Sri Lanka’s physical-person definition and silence on a specific computer-generated rule counsel against judicially inventing an equivalent statutory fiction.71

A deemed-authorship rule can encourage investment by supplying a rights holder when no human can fairly claim to have fashioned the expressive details.72 But this convenience comes at a cost: it may reward capital ownership and access to computing power rather than creative activity, confer excessively broad exclusivity over abundant machine outputs, and obscure the difference between owning the instrument and creating the work.73 As the United Kingdom’s subsequent public consultation demonstrates, the continuing suitability of legacy computer-generated-work protection is a live policy issue rather than a consensus already reached.74

Comparative approach III: European originality

The Court of Justice of the European Union has developed a criterion of the “author’s own intellectual creation”. In Infopaq, the court connected copyright subject matter with original expressive choices, while Painer demonstrated that a photographer’s decisions regarding background, pose, lighting and framing can make a technology-assisted photograph an original work.75 Cofemel and Brompton Bicycle further emphasised free and creative choices and the expression of the author’s personality, including where functional constraints exist.76 These judgments concern their respective factual contexts; they do not establish a bespoke EU-wide generative AI ownership rule.77

For Sri Lanka, the usefulness of this jurisprudence lies in the distinction between creative determination and technically constrained execution. Where a person makes choices that meaningfully shape the form of expression and the software executes them, AI assistance need not displace authorship.78 Where the system determines the decisive visual or linguistic particulars despite the user’s broad thematic instructions, the connection is weaker.79 The test must consider what creative freedom the human actually exercised, not merely how polished, original-looking or labour-intensive the output appears.80 Complexity is not itself proof of a legally relevant human origin.

The European position also highlights the need to keep authorship separate from upstream training disputes. The Information Society Directive establishes exclusive rights, and the Digital Single Market Directive contains text-and-data-mining exceptions subject to conditions.81 Those rules affect uses of prior works in building and exploiting systems.82 They are not answers to who authored a new output. A claimant may fail to demonstrate copyright in an autonomous output but still have infringed a photographer’s original work in the course of generating it; alternatively, an original human-authored AI-assisted work might be independently protectable while incorporating unlawfully borrowed material.

Where should the line be drawn?

A workable line requires consideration of at least five common creative activities. First, a user enters a short prompt, such as “paint an ancient Sri Lankan harbour at sunset”, then publishes the first image.83 The user has communicated subject matter and perhaps mood, but the model has supplied composition, figures, colour relations and countless other details. The prompt itself may be protected if original, but the user has a weak claim to have authored the visual expression.84 Neither artistic quality nor the user’s intention to publish resolves the problem.85 The copyright inquiry must focus on which elements the person actually created.

Second, a sophisticated prompt may specify camera angle, placement of figures, colour palette, precise text, foreground relationships and multiple constraints. This involvement strengthens the argument for human determination, particularly if the finished output faithfully realises an individually conceived composition.86 Yet the number of words in the prompt cannot provide an automatic legal threshold.87 A thousand-word instruction may still leave aesthetic details to stochastic generation, whereas a short instruction controlling a specific expressive transformation might be significant.88 Evidence of realised creative control is more persuasive than a token count or the mere existence of iterative commands.

Third, a user generates fifty images, chooses one and publishes it unchanged. Selection demonstrates taste, effort and perhaps editorial judgment, but copyright in the selected image is not automatically created by the act of choosing.89 A genuine originality claim may arise if that image is part of a carefully authored selection and arrangement of several works, producing a new expressive structure as a compilation.90 In such a case protection should relate to that original structure, not retrospectively monopolise the unaltered component images.91 Sri Lanka’s section 7 supplies a particularly appropriate statutory basis for this form of partial protection.

Fourth, a human editor substantially repaints generated imagery, rewrites sentences, inserts original dialogue, changes musical phrases or designs a distinctive sequence of shots. Those human-created expressive elements can qualify under section 6 if original, even if the initial synthetic material does not.92 The amount of editing is relevant but not decisive.93 Routine spelling changes, colour corrections or mechanical filters may not be creative, while substantial restructuring can be.94 A careful judgment should identify what expression was added, what was merely preserved from the AI system, and whether the contributions are separable for the purpose of the rights claimed.

Fifth, a researcher drafts a fully original legal argument and asks AI to improve grammar, reorganise headings or test objections. Authorship of the original argument and its expression ordinarily remains with the researcher to the extent those features survive; the tool does not become a joint author.95 However, if the system rewrites the argument into markedly different substantive text and the researcher accepts it with minimal intellectual involvement, the claim to authorship of particular passages becomes more complicated.96 Authorship cannot be established simply by saying “I approved the final answer”.97 Editorial responsibility, academic integrity and copyrightability also remain distinct questions.

The decisive distinction is between choosing an outcome from machine-made alternatives and creatively determining expression in the resulting work. Selection and arrangement may be independently original; original modifications may be protected; precisely directed generation may sometimes embody the user’s sufficiently specified expressive design.98 But ownership of the full unmodified output should not follow from a causal chain alone.99 It is more coherent to accept partial and layered protection than to force every creative project into the binary of wholly human or wholly machine-made.100 This approach also aligns remedies with the contribution actually proven.

A proposed Sri Lankan test of human creative contribution

This article proposes a Human Creative Contribution Test (HCCT), to be applied to the specific expression for which protection is claimed. The first stage is attribution: the claimant must identify one or more physical persons who allegedly originated the relevant expression.101 Naming an AI service, company or algorithm as sole author fails under section 5.102 Naming a human is necessary but insufficient, because initial ownership and presumed attribution cannot replace evidence of creation.103 If an employer or commissioner asserts economic rights, the court should still determine who supplied original authorship and then apply section 14 or the relevant contractual provisions.

The second stage is expressive specification. A court should identify the claimed original features of the work (such as arrangement of arguments, wording, visual composition, melodic progression or selection of components) and ask which features the human conceived, selected, directed or altered.104 General ideas, styles, genres and subject matter must be filtered out under section 8. The exercise need not reconstruct every algorithmic step. Rather, it requires an intelligible account of the claimant’s creative contribution at a level of specificity appropriate to the work.105 Unsupported claims of “artistic vision” should not be enough when no corresponding authored expression can be identified.

The third stage is effective control and causation. The question is whether the human choices materially shaped the final form, or merely set a process in motion and later accepted its result.106 Evidence may include original sketches, drafts, prompts, negative prompts, version histories, layer files, marked changes, written plans and comparisons between initial and final outputs.107 The inquiry remains qualitative: control can be shared between human and system, and a human need not predict every pixel to author an image.108 Yet the closer an output lies to the model’s independent completion of an open-ended request, the weaker the claim to authored expressive details.

The fourth stage is originality. Once human contribution has been isolated, it must satisfy section 6’s requirement of original intellectual creation as interpreted by Sri Lankan case law.109 A purely automatic transcription, a commonplace instruction or a banal reformatting step will not necessarily cross that threshold. Conversely, authorship should not be denied because the creative work is commercially modest or visually simple.110 The court should ask whether there is independent creative activity in expression, not whether a human used a particular traditional method.111 This guards against both technology discrimination and the creation of exclusivity without genuine authorship.

The fifth stage is protection and allocation. Where the evidence proves only original human editing, selection or arrangement, protection should be confined to those elements or to their original combination.112 The court should next address pre-existing rights, commissioned-work rules, employment relationships, written assignments and remedies.113 A declaration describing the protected human contribution will often be more precise than a declaration that the complete AI-mediated file is copyrighted.114 In doubtful cases, a claimant should not gain rights in objectively unproven AI-determined components by hiding them within a larger original work.

This structured test would not require a separate new originality standard for AI. Instead, it translates the existing physical-person requirement and originality threshold into an evidential method suited to modern production.115 It avoids a rigid rule that all prompts are legally irrelevant, because narrowly specified and effectively realised instructions may reflect authored expression.116 Equally, it avoids treating elaborate prompting or mere economic investment as sufficient in every case.117 Crucially, the test is neutral as between text, images, music and computer-generated designs, although the relevant forms of expressive control may differ across media.118

Ownership, moral rights and infringement

In Sri Lanka the initial author and the eventual economic rights holder may be different persons.119 Section 14 deals expressly with joint authorship, collective works, employment, commissions and audiovisual productions. A publishing house can therefore hold economic rights in qualifying human-written portions through the applicable statutory or contractual route without being a physical author of each passage.120 But the existence of an employment contract, AI subscription licence or commercial invoice cannot create copyright where no qualifying human expression has been shown.121 Nor does a platform’s contractual declaration that users “own outputs” bind third parties on the separate issue of copyright subsistence.

Moral rights intensify the difficulty of fictitious machine authorship. Section 10 protects attribution and reputation interests that make intelligible sense in relation to a human creator.122 An AI cannot presently suffer injury to personal honour or exercise a right to object to derogatory treatment in the sense contemplated by the statute.123 Equally, a user should not describe themselves as having authored wholly generated passages merely to obtain the social or reputational advantage of attribution.124 Honesty about tool use is an ethical consideration, while the legal question remains whether the claimant actually supplied the expression.

The training and infringement debates should be addressed without displacing the article’s core question. AI models may be trained on protected texts, illustrations and recordings, prompting arguments about reproduction, fair use, licensing and exceptions.125 Getty Images v. Stability AI illustrates how complex those separate claims can become, and its procedural and substantive holdings must be read with care rather than characterised as a universal rule about AI outputs.126 In Sri Lanka section 11 contains a fair-use framework, but the permissible training uses and the authorship of generated material are different legal questions that may arise in the same commercial dispute.127

Joint authorship should likewise be approached with precision. The Act contemplates contributions by two or more authors, but the mere presence of an AI system does not produce a qualifying second human author.128 A programmer, prompt engineer and illustrator could be joint or separate human authors only if their actual creative contributions meet the statutory requirements and the nature of their collaboration warrants that classification.129 Model ownership does not itself create joint authorship; neither does providing a generic software service.130 Courts should distinguish intellectual contribution, contractual control and downstream assignment rather than using “collaboration” as an undifferentiated label.

Implications for Sri Lankan creators and institutions

For creative industries in Sri Lanka, uncertainty can impair licensing, investment and enforcement.131 A designer who relies on AI to generate a background image and then independently designs typography and a logo may have enforceable rights in the original layout or lettering but not necessarily exclusive rights in every generated background feature. A filmmaker who combines original screenplay, direction and editing with AI-created scene elements should identify the rights in the human-authored components and audiovisual work under the relevant rules.132 Accurate rights statements can reduce disputes without suppressing useful technological assistance.133

For universities, research institutes and publishers, the issue is particularly sensitive. AI can assist with proofreading, translation, literature searches and drafting, but a scholar’s claim to the authorship of a legal article should be grounded in independent intellectual formulation and meaningful control over the published expression.134 Copyright attribution is not a substitute for compliance with academic misconduct rules, accurate citation or verification of authorities. A journal should therefore distinguish permitted assistance from disclosure requirements and from proof of originality; those standards serve overlapping but not identical purposes.135

Sri Lanka’s linguistic environment adds another dimension.136 Sinhala and Tamil creators may employ generative systems to translate, adapt or edit their writing across languages. Human revision can be intellectually demanding and sometimes original, but merely accepting an automatic translation does not establish authorship of every translated phrase.137 Section 7 can protect qualifying original transformations, subject to rights in the source work.138 A balanced rule would encourage multilingual access while protecting both the author of the underlying work and the translator’s demonstrable creative contributions.

Small businesses and independent artists may face a greater evidential burden than large firms able to preserve logs and hire lawyers. The proposed framework should therefore accept a variety of reliable evidence, not demand proprietary software records unavailable to the user. Drafts, testimony, saved iterations, contemporaneous notes and visual comparisons may suffice. Courts should also remain alert to confidentiality and data-protection considerations when requesting prompt histories containing personal or commercially sensitive information.139 The objective is proof of creative contribution, not compulsory disclosure of every private interaction with an AI system.

Reform options and safeguards

First, the National Intellectual Property Office could publish non-binding guidance explaining the continuing importance of sections 5–8 and illustrating different levels of human participation. Examples should encompass short prompts, detailed direction, original editing, compilations, AI-assisted translation and fully autonomous output.140 The guidance should clearly distinguish the existence of copyright from ownership and contract rights.141 Such a measure would assist creators immediately without pretending to settle issues that properly require judicial determination or legislation. Because Sri Lanka does not make ordinary copyright subsistence depend on registration, a new registration requirement should not be smuggled into guidance.142

Second, disclosure could be calibrated to circumstances. A claimant who seeks an injunction or damages for an AI-mediated work should identify, when genuinely contested, the human-authored elements relied upon and material AI involvement relevant to those claims.143 Publishers might adopt standard declaration fields for production workflows. However, a duty to disclose every trivial spelling correction or software suggestion would be disproportionate and difficult to enforce.144 Disclosure should improve the identification of protected expression rather than stigmatise any use of new tools. The American Copyright Office’s registration practice offers useful comparative illustrations but should not be copied mechanically.145

Third, legislation could expressly clarify partial protection for original human contributions in AI-assisted material, while preserving the definition of a human author.146 A carefully drafted provision might state that use of an automated system does not of itself prevent protection of independently created human expression, selection, arrangement or modification; nor does the act of instructing a system, without sufficient creative determination of expression, alone confer authorship of its generated output.147 This would formalise the proposed test while remaining consistent with the Act’s treatment of original works and derivative compilations.148

Fourth, Sri Lanka should be cautious about importing the United Kingdom’s deeming provision for works lacking any human author. Before creating a new proprietary entitlement, lawmakers should demand empirical evidence that unprotected autonomous outputs cause a socially significant investment failure.149 An alternative could be limited contractual protection, trade secrecy where appropriate, or competition-based remedies without constructing artificial copyright authorship.150 The opposing concern, namely the concentration of rights in model providers or industrial-scale generators, deserves equal attention.151 Any reform should specify beneficiaries, scope, duration, exceptions and the relationship to ordinary human copyright.

Finally, the reform process should involve writers, artists, academics, technology firms, publishers, cultural institutions and public-interest organisations, including Sinhala- and Tamil-language creators. International obligations under Berne and TRIPS provide important constraints and guarantees, but they do not themselves supply an exhaustive generative AI authorship code.152 Comparative experience remains unsettled and fast-moving.153 A principled Sri Lankan response should therefore preserve flexibility in adjudication, resist premature monopolisation, and focus on verifiable creative contribution rather than a contested philosophical account of whether a machine truly “creates”.154

Conclusion

The dilemma of authorship in the age of AI is not resolved by a simple assertion that a human clicked the button or that a machine produced the pixels. Sri Lankan law already provides the foundation for a more careful answer: a qualifying author must be a physical person and a protected work must embody original intellectual creation.155 Those requirements exclude attributing legal authorship to an AI system as such, while leaving substantial room for the protection of expressive human contributions realised through technological assistance.156 Existing provisions on derivative works, ownership and moral rights reinforce the need to identify the contribution with precision.157

The proposed Human Creative Contribution Test draws the line through attribution, expressive specification, effective control, originality and allocation. The result may be full copyright in a human-directed work, partial protection of editing or arrangement, or no copyright in an autonomously generated component.158 This conclusion does not deprive human creators of legitimate rights, nor does it grant a monopoly merely because a costly and complex system has produced an appealing result.159 The appropriate measure of authorship is the legally relevant human creativity embodied in expression.160 For Sri Lanka, clarifying that principle is a stronger and more defensible response than deeming every AI user an author.

*****

Footnotes

1. See Jane C. Ginsburg & Luke Ali Budiardjo, Authors and Machines, 34 Berkeley Tech. L.J. 343 (2019); Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 60 (1884) (photographer’s posing of the subject, arrangement of costume and accessories, and disposition of light and shade).

2. See generally Pamela Samuelson, Allocating Ownership Rights in Computer-Generated Works, 47 U. Pitt. L. Rev. 1185 (1986).

3. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability 18 (2025) [hereinafter Part 2 Report].

4. Part 2 Report, supra note 3, at iii.

5. See Daniel J. Gervais, The Machine as Author, 105 Iowa L. Rev. 2053 (2020).

6. Intellectual Property Act, No. 36 of 2003, § 9(1)–(2) (Sri Lanka).

7. Part 2 Report, supra note 3, at iii.

8. See Letter from Robert J. Kasunic, Assoc. Reg. of Copyrights & Dir. of the Off. of Registration Pol’y & Prac., U.S. Copyright Off., to Van Lindberg, Taylor English Duma LLP, at 1 (Feb. 21, 2023) (re Zarya of the Dawn, Registration # VAu001480196) [hereinafter Zarya of the Dawn Letter].

9. Intellectual Property Act §§ 5, 6(1) (Sri Lanka).

10. See Ruwan Fernando, Intellectual Property Adjudication in Sri Lanka 195 (WIPO & U.N. Dev. Programme 2025) (Intellectual Property Benchbook Series).

11. See Intellectual Property Act §§ 5, 14(1), (4) (Sri Lanka).

12. See generally Fernando, supra note 10, at 194–98.

13. Copyright, Designs and Patents Act 1988, c. 48, § 9(3) (UK); The Copyright Act, 1957, § 2(d)(vi), No. 14, Acts of Parliament, 1957 (India); Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), cert. denied, No. 25-449 (U.S. Mar. 2, 2026); Case C-145/10, Painer v. Standard VerlagsGmbH, EU:C:2011:798, ¶¶ 88–94 (Dec. 1, 2011).

14. See generally Jane C. Ginsburg, The Concept of Authorship in Comparative Copyright Law, 52 DePaul L. Rev. 1063 (2003).

15. Cf. Part 2 Report, supra note 3, at iii (distinguishing the use of AI tools to assist human creativity from their use to stand in for it).

16. Zarya of the Dawn Letter, supra note 8, at 1.

17. See Part 2 Report, supra note 3, at iii, 24–27.

18. See Intellectual Property Act §§ 5, 6(1), 9, 14 (Sri Lanka).

19. See Intellectual Property Act §§ 9(2), 14(3)–(4) (Sri Lanka).

20. See Intellectual Property Act § 5 (Sri Lanka) (defining the author as “the physical person who has created the work”); id. § 14(4) (vesting economic rights in a commissioner, not authorship).

21. Intellectual Property Act § 5 (Sri Lanka).

22. Fernando, supra note 10, at 194–95.

23. Fernando, supra note 10, at 194; Intellectual Property Act § 14(3)–(4) (Sri Lanka).

24. Intellectual Property Act § 6(1)–(2) (Sri Lanka).

25. Fernando, supra note 10, at 197, 199.

26. Fernando, supra note 10, at 195–97.

27. Intellectual Property Act § 7(1) (Sri Lanka).

28. See L. Lionel Halwathura v. Sirisumana Godage, Case No. HC/Civil/60/2006(3) (Commercial High Ct., Colombo 2016) (unreported) (Sri Lanka), discussed in Fernando, supra note 10, at 197–98 (derivative work protected only where it shows substantial, not trivial, variation from the underlying work).

29. Intellectual Property Act § 7(2) (Sri Lanka).

30. Intellectual Property Act § 8(a) (Sri Lanka).

31. Part 2 Report, supra note 3, at 13, 18.

32. See Designers Guild Ltd. v. Russell Williams (Textiles) Ltd. [2000] 1 WLR 2416 (HL) (appeal taken from Eng.); Intellectual Property Act § 8(a) (Sri Lanka).

33. Intellectual Property Act §§ 9(1), 10(1) (Sri Lanka).

34. Intellectual Property Act §§ 13(1), 14 (Sri Lanka).

35. Intellectual Property Act §§ 15(1), 16 (Sri Lanka).

36. Wijesinghe Mahanamahewa v. Austin Canter, [1986] 2 Sri LR 154 (CA) (Sri Lanka); see Fernando, supra note 10, at 196.

37. Vasantha Obeysekera v. Anthony Christopher Alles, CA No. 730/92(F), at 10–12 (Ct. App. 2000) (unreported) (Sri Lanka), discussed in Fernando, supra note 10, at 196.

38. Fernando v. Gamlath, [2011] 1 Sri LR 273 (SC) (Sri Lanka); Univ. of London Press Ltd. v. Univ. Tutorial Press Ltd. [1916] 2 Ch 601, 608–09 (Eng.).

39. See Part 2 Report, supra note 3, at 20–21.

40. See Intellectual Property Act §§ 6(1)(a), 7(1)(b) (Sri Lanka).

41. Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, 345, 352–54 (1991); CCH Canadian Ltd. v. Law Soc’y of Upper Can., [2004] 1 S.C.R. 339, para. 16 (Can.).

42. Fernando, supra note 10, at 194–95.

43. Fernando, supra note 10, at 197.

44. Fernando, supra note 10, at 195.

45. Compare Intellectual Property Act § 5 (Sri Lanka), with Copyright, Designs and Patents Act 1988, § 9(3) (UK).

46. Fernando, supra note 10, at 195.

47. See Halwathura, supra note 28; CCH, [2004] 1 S.C.R. 339, para. 16 (trivial changes, such as altering a font, do not suffice).

48. Cf. Case C-5/08, Infopaq Int’l A/S v. Danske Dagblades Forening, EU:C:2009:465, ¶¶ 37, 44–45 (July 16, 2009) (locating originality in the author’s choice, sequence and combination of words).

49. Feist, 499 U.S. at 345.

50. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190, 16,193 (Mar. 16, 2023) [hereinafter Registration Guidance].

51. Part 2 Report, supra note 3, at iii, 18.

52. Part 2 Report, supra note 3, at iii, 24–27.

53. Zarya of the Dawn Letter, supra note 8, at 1.

54. Letter from the Review Bd., U.S. Copyright Off., to Tamara Pester, Esq., at 1 (Sept. 5, 2023) (re Second Request for Reconsideration for Refusal to Register Théâtre D’opéra Spatial, SR # 1-11743923581) [hereinafter Théâtre D’opéra Spatial Decision].

55. Cf. Fernando, supra note 10, at 194–95 (referring to United States decisions by way of comparison while noting that the question remains undecided in Sri Lanka).

56. Thaler, supra note 13 (declining to resolve line-drawing disputes over AI-assisted works because the application named the machine as the sole author).

57. See Thaler v. Perlmutter, 687 F. Supp. 3d 140 (D.D.C. 2023), aff’d, 130 F.4th 1039 (D.C. Cir. 2025).

58. Burrow-Giles, 111 U.S. at 60.

59. Compare Intellectual Property Act § 5 (Sri Lanka), with 17 U.S.C. §§ 102(a)–(b), 201(a).

60. Intellectual Property Act §§ 5, 6(2) (Sri Lanka); Fernando, supra note 10, at 199.

61. Cf. Registration Guidance, 88 Fed. Reg. at 16,193.

62. Copyright, Designs and Patents Act 1988, § 9(3) (UK).

63. Copyright, Designs and Patents Act 1988, § 178 (UK).

64. Andres Guadamuz, Artificial Intelligence and Copyright, WIPO Mag., Oct. 2017.

65. See Ginsburg & Budiardjo, supra note 1.

66. Nova Prods. Ltd. v. Mazooma Games Ltd. [2007] EWCA (Civ) 219 (Eng.).

67. Cf. Part 2 Report, supra note 3, at 30 (noting the lack of any case law applying the UK provision to AI).

68. See Nova Prods., supra note 66.

69. The Copyright Act, 1957, § 2(d)(vi) (India).

70. See Eastern Book Co. v. D.B. Modak, (2008) 1 SCC 1 (India) (requiring a minimal degree of creativity for originality).

71. Intellectual Property Act § 5 (Sri Lanka); but see Fernando, supra note 10, at 195.

72. See Robert C. Denicola, Ex Machina: Copyright Protection for Computer-Generated Works, 69 Rutgers U. L. Rev. 251 (2016); see also Ryan Abbott & Elizabeth Rothman, Disrupting Creativity: Copyright Law in the Age of Generative Artificial Intelligence, 75 Fla. L. Rev. 1141 (2023); Annemarie Bridy, Coding Creativity: Copyright and the Artificially Intelligent Author, 5 Stan. Tech. L. Rev. 1 (2012).

73. See Dep’t for Sci., Innovation & Tech. & Dep’t for Culture, Media & Sport, Report on Copyright and Artificial Intelligence 14, ¶ 46 (Mar. 2026) (UK) [hereinafter UK Report] (observing that protection for computer-generated works “departs from the core rationale for copyright, which is to encourage and reward human creativity”); Gervais, supra note 5.

74. Intellectual Property Office, Copyright and Artificial Intelligence (Consultation, Dec. 17, 2024) (UK); UK Report, supra note 73, at 14, ¶ 47 (reporting that most respondents supported removing protection for works created solely by AI).

75. Infopaq, EU:C:2009:465, ¶¶ 37, 45; Painer, EU:C:2011:798, ¶¶ 90–92.

76. Case C-683/17, Cofemel – Sociedade de Vestuário SA v. G-Star Raw CV, EU:C:2019:721, ¶¶ 30–31 (Sept. 12, 2019); Case C-833/18, SI & Brompton Bicycle Ltd. v. Chedech/Get2Get, EU:C:2020:461 (June 11, 2020).

77. See Part 2 Report, supra note 3, at 28–29 (surveying international approaches, including that of the European Union).

78. See Painer, EU:C:2011:798, ¶¶ 90–92; Burrow-Giles, 111 U.S. at 60.

79. Part 2 Report, supra note 3, at 18.

80. See Cofemel, EU:C:2019:721, ¶¶ 30–31; Feist, 499 U.S. at 352–54.

81. 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, arts. 2–4, 2001 O.J. (L 167) 10; 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, arts. 3–4, 2019 O.J. (L 130) 92.

82. See Directive (EU) 2019/790, supra note 81, arts. 3–4; Getty Images (US) Inc. v. Stability AI Ltd. [2025] EWHC 2863 (Ch) (Eng.).

83. See Part 2 Report, supra note 3, at 18.

84. Part 2 Report, supra note 3, at 13, 18.

85. See Intellectual Property Act § 6(2) (Sri Lanka) (protection “irrespective of ... content, quality and purpose”).

86. See Ginsburg & Budiardjo, supra note 1.

87. See Part 2 Report, supra note 3, at 18–20.

88. See Part 2 Report, supra note 3, at 18, 22–24.

89. Part 2 Report, supra note 3, at 21.

90. See Zarya of the Dawn Letter, supra note 8, at 1; Feist, 499 U.S. at 348; Intellectual Property Act § 7(1)(b) (Sri Lanka).

91. See Halwathura, supra note 28, discussed in Fernando, supra note 10, at 198 (copyright in a derivative work extends only to the new material added).

92. Intellectual Property Act § 6(1) (Sri Lanka); Part 2 Report, supra note 3, at iii, 24–27.

93. See Halwathura, supra note 28.

94. See CCH, [2004] 1 S.C.R. 339, para. 16; Feist, 499 U.S. at 345.

95. See Intellectual Property Act § 5 (Sri Lanka) (defining an “author” as a physical person and a “work of joint authorship” as one to which two or more authors have contributed).

96. See Part 2 Report, supra note 3, at iii.

97. See Part 2 Report, supra note 3, at 21 (authorship by adoption does not in itself support a claim to AI-generated output).

98. See Part 2 Report, supra note 3, at iii.

99. See Gervais, supra note 5.

100. See Zarya of the Dawn Letter, supra note 8, at 1; cf. Samuelson, supra note 2.

101. Intellectual Property Act § 5 (Sri Lanka).

102. Id.; cf. Thaler, supra note 13.

103. See Intellectual Property Act § 15(1) (Sri Lanka) (presumption of authorship applies “in the absence of proof to the contrary”).

104. See Infopaq, EU:C:2009:465, ¶ 45; Feist, 499 U.S. at 348; cf. Intellectual Property Act § 8(a) (Sri Lanka); Gamlath, supra note 38 (copyright protects the expression of thought, not ideas).

105. Cf. Registration Guidance, 88 Fed. Reg. at 16,193 (requiring applicants to describe the human author’s contribution).

106. Part 2 Report, supra note 3, at 18–20; see also Gervais, supra note 5.

107. Cf. Théâtre D’opéra Spatial Decision, supra note 54, at 2 (recording the applicant’s account of his prompting and subsequent editing).

108. See Burrow-Giles, 111 U.S. at 60; Painer, EU:C:2011:798, ¶¶ 90–92.

109. Intellectual Property Act § 6(1) (Sri Lanka); Mahanamahewa, supra note 36; Gamlath, supra note 38.

110. Intellectual Property Act § 6(2) (Sri Lanka); Feist, 499 U.S. at 345.

111. Fernando, supra note 10, at 197.

112. See Halwathura, supra note 28; Zarya of the Dawn Letter, supra note 8, at 1.

113. Intellectual Property Act §§ 7(2), 14(4), 16 (Sri Lanka).

114. Cf. Zarya of the Dawn Letter, supra note 8, at 1 (replacing the registration with one covering only the expressive material the applicant created).

115. Intellectual Property Act §§ 5, 6(1) (Sri Lanka).

116. See Part 2 Report, supra note 3, at 22–24.

117. Part 2 Report, supra note 3, at 20; Feist, 499 U.S. at 352–54.

118. See Intellectual Property Act § 6(1)–(2) (Sri Lanka).

119. Intellectual Property Act § 14 (Sri Lanka).

120. Intellectual Property Act § 14(2)–(5) (Sri Lanka).

121. See Intellectual Property Act §§ 6(1), 16(1) (Sri Lanka) (an owner may license or assign only the economic rights in an existing work).

122. Intellectual Property Act § 10(1) (Sri Lanka).

123. Intellectual Property Act § 10(1)(c), (2) (Sri Lanka).

124. Cf. Registration Guidance, 88 Fed. Reg. at 16,193.

125. See Directive (EU) 2019/790, supra note 81, arts. 3–4; Getty Images, supra note 82.

126. Getty Images, supra note 82.

127. Intellectual Property Act § 11 (Sri Lanka); cf. Google LLC v. Oracle Am., Inc., 593 U.S. 1 (2021) (applying the corresponding United States four-factor fair-use test, a question distinct from authorship).

128. Intellectual Property Act § 5 (Sri Lanka) (definition of “work of joint authorship”).

129. Intellectual Property Act §§ 5, 14(2) (Sri Lanka).

130. See generally Ginsburg & Budiardjo, supra note 1.

131. See generally World Intellectual Property Organization, Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence, WIPO/IP/AI/2/GE/20/1 REV (May 21, 2020).

132. Intellectual Property Act § 14(5) (Sri Lanka).

133. See Intellectual Property Act § 15(1) (Sri Lanka).

134. Intellectual Property Act § 5 (Sri Lanka); Fernando, supra note 10, at 197.

135. Cf. Registration Guidance, 88 Fed. Reg. at 16,193 (disclosure for the distinct purpose of copyright registration).

136. Const. art. 18 (Sri Lanka).

137. See Halwathura, supra note 28 (trivial variation insufficient).

138. Intellectual Property Act § 7(1)(a), (2) (Sri Lanka).

139. See Personal Data Protection Act, No. 9 of 2022 (Sri Lanka).

140. Intellectual Property Act §§ 5–8 (Sri Lanka).

141. Intellectual Property Act §§ 14, 16 (Sri Lanka).

142. Intellectual Property Act § 6(2) (Sri Lanka); Fernando, supra note 10, at 199.

143. Cf. Registration Guidance, 88 Fed. Reg. at 16,193.

144. Cf. id. (requiring disclosure only of AI-generated content that is more than de minimis).

145. See generally Part 2 Report, supra note 3, at iii.

146. Cf. UK Report, supra note 73, at 14, ¶ 47.

147. Cf. Part 2 Report, supra note 3, at iii.

148. Intellectual Property Act §§ 6, 7 (Sri Lanka).

149. See UK Report, supra note 73, at 14, ¶ 47; id. at 110–11, ¶¶ 46–52 (noting the lack of evidence of the value of computer-generated-works protection).

150. Cf. Part 2 Report, supra note 3, at iii, 36 (concluding that the case has not been made for additional copyright or sui generis protection).

151. Cf. WIPO, supra note 131.

152. Berne Convention for the Protection of Literary and Artistic Works arts. 2, 5, 6bis, Sept. 9, 1886, as revised at Paris July 24, 1971, and amended Sept. 28, 1979; Agreement on Trade-Related Aspects of Intellectual Property Rights arts. 9–10, Apr. 15, 1994, Marrakesh Agreement Establishing the World Trade Organization, Annex 1C, 1869 U.N.T.S. 299.

153. See Part 2 Report, supra note 3, at 28–31.

154. Cf. WIPO, supra note 131.

155. Intellectual Property Act §§ 5, 6(1) (Sri Lanka).

156. Fernando, supra note 10, at 194–95.

157. Intellectual Property Act §§ 7, 10, 14 (Sri Lanka).

158. See Zarya of the Dawn Letter, supra note 8, at 1.

159. See Feist, 499 U.S. at 352–54.

160. See Ginsburg & Budiardjo, supra note 1.

How to cite

Menaka Harankaha, The Dilemma of Authorship in the Age of Artificial Intelligence: Drawing the Line Between AI-Generated and AI-Assisted Works in Copyright Law from a Sri Lankan Perspective, 9 Int'l J.L. Mgmt. & Human. 1166 (2026), https://doi.org/10.63108/IJLMH.13303.

Cite this article

Menaka Harankaha, The Dilemma of Authorship in the Age of Artificial Intelligence: Drawing the Line Between AI-Generated and AI-Assisted Works in Copyright Law from a Sri Lankan Perspective, 9 Int'l J.L. Mgmt. & Human. 1166 (2026), https://doi.org/10.63108/IJLMH.13303.
Chat with us