Who Owns the Dress?AI-Generated Fashion Designs and the Human Authorship Crisis

Abstract

The proliferation of generative artificial intelligence in fashion design confronts copyright law with a challenge that is not merely technical but foundational. When a garment silhouette, textile pattern, or couture concept emerges from sustained dialogue between a human prompt-engineer and a diffusion model, the question of who, if anyone, holds the resulting copyright attracts competing claims from individual designers, fashion houses, AI developers, and, in some accounts, the AI system itself. This article approaches that question through doctrinal, comparative, and policy analysis, drawing on the authorship jurisprudence of the United States, the United Kingdom, the European Union, and India. The orthodox human-authorship requirement is, this article argues, both constitutionally defensible and philosophically coherent; the problem is not the requirement itself but the inadequacy of existing frameworks for applying it to creative processes in which the human contribution operates upstream of, rather than upon, the final aesthetic artefact. Neither categorical denial of protection for AI outputs nor wholesale assignment of rights to AI developers adequately reflects the creative economy’s actual structure. The article proposes instead a contribution-weighted entitlement framework, calibrated by the degree of substantive human creative control exercised at each stage of the design process, as the most doctrinally rigorous and commercially workable response to an authorship crisis the law has not yet found the vocabulary to address.

I. Introduction

When Valentino’s creative director described his studio’s experimental use of diffusion-model software as “a conversation with a mirror that knows more than you do,” he articulated one of the most consequential unresolved questions in contemporary intellectual property law. Who owns what the mirror shows remains, as of this writing, genuinely open, and the stakes are far higher than the fashion industry has publicly acknowledged.

Fashion design already occupies contested IP ground. In the United States, garment design receives only attenuated copyright protection, and the Copyright Office has historically struggled to separate protectable aesthetic elements from functional dimensions.¹ In the European Union, registered Community designs coexist with national copyright regimes that remain divided, most visibly between French and German approaches, on what counts as an “author’s own intellectual creation.” In the United Kingdom, post-Brexit design law preserves much of the EU acquis while facing the additional complication of the computer-generated works provision of the Copyright, Designs and Patents Act 1988, which has received little of the judicial scrutiny its commercial importance now demands.² The arrival of AI authorship claims threatens frameworks that were under pressure before any generative model produced its first textile print.

The question this article sets out to answer is deceptively simple: to what extent should copyright law recognise human authorship in AI-generated fashion designs, and how should ownership be allocated when creative contributions are distributed among users, developers, fashion houses, and AI systems? The dominant scholarly positions—categorical denial of protection on the one hand, wholesale extension of the work-for-hire doctrine on the other—both fail to capture the genuine complexity of how AI-assisted design is actually practised. This article proposes a contribution-weighted entitlement framework that locates protectability in the measurable degree of substantive human creative control exercised at each stage of the design process, a framework that is both doctrinally coherent under existing authority and, with modest legislative intervention, commercially administrable. Part II sets out the conceptual background; Part III analyses the competing ownership claims; Part IV examines three leading cases; Part V develops the proposed framework; and Part VI concludes.

II. Background and Conceptual Framework

A. The Historical Architecture of Copyright Authorship

Copyright law’s commitment to human authorship is neither accidental nor arbitrary. The Statute of Anne 1710 vested rights in “authors,” a term whose humanist connotations have never been seriously dislodged. The Anglo-American expressionist tradition protects the imprint of an individual personality upon a cultural artefact; the United States Supreme Court’s “creative spark” requirement demands that protected works reflect, however modestly, the author’s own intellectual conception—a formulation elastic enough to accommodate photography and film, though each extension required fresh doctrinal labour at the margins.³ The Continental droit d’auteur framework goes further, treating the author not merely as a hook for vesting rights but as the very source of the value the law protects. Berne Convention Article 2 reflects a deliberate ambiguity, leaving signatory states latitude to define authorship domestically—latitude that has become increasingly consequential as AI systems demonstrate capabilities once treated as definitionally human.⁴

What the AI authorship debate has thrown into sharp relief is that the human-authorship requirement has always been doing double duty: it operates simultaneously as a threshold eligibility rule and as a structural assumption about the nature of creative value. That conflation served tolerably well for three centuries, but it begins to crack when a sophisticated human designer can deploy a generative AI system to produce outputs with genuine aesthetic originality without making the “original” choices that copyright doctrine requires.

B. Originality and the Threshold of Protection

Feist Publications, Inc v Rural Telephone Service Co established that originality demands both independent creation and at least some minimal degree of creativity, excluding purely mechanical productions.⁵ The EU Court of Justice in Infopaq and Football Dataco similarly insists that copyright protects only works reflecting the author’s “free and creative choices” rather than technical constraint.⁶ Taken together, these formulations press hard against AI-generated works. The user who prompted the system made choices about the prompt, not the output. The developer made choices about architecture and dataset, not the aesthetic features of any particular generated design. The conceptual gap between human creative act and AI-generated artefact is not merely a matter of degree; it is structural, and it resists closure by doctrinal analogy alone.

A further complication, underexplored in the literature, runs deeper still. The training datasets on which generative fashion AI is built aggregate the creative labour of thousands of designers whose choices about silhouette, colour, proportion, and historical reference are encoded in the model’s weights. The AI-generated output reflects a statistical synthesis of human creativity at a population level, even if no individual designer’s choices are directly reproduced. Whether that diffuse human input is sufficient to ground an originality claim, and in whom it might vest, are questions existing doctrine is simply not equipped to answer.

C. Fashion Design Protection: A Layered and Contested Framework

In the United States, copyright protection for fashion designs depends on the separability test articulated in Star Athletica, LLC v Varsity Brands, Inc.⁷ The practical scope is narrow: three-dimensional garment designs rarely satisfy the test, and the United States has never enacted a specific fashion design statute—an omission that looks increasingly anomalous as the industry moves into AI-generated design at scale. European law offers considerably richer protection under Directive 98/71/EC and Regulation (EC) 6/2002,⁸ with France’s principle of the unité de l’art remaining the most expansive approach because it least requires the aesthetic/functional separation that bedevils American doctrine. The UK framework faces the additional complication of section 9(3) of the 1988 Act—a provision of particular comparative significance whose implications for AI-generated works have never been definitively resolved by any English court.⁹ The introduction of AI-generated designs does not merely complicate these frameworks; it exposes the extent to which they were already contingent, contested, and dependent on unstated assumptions about the nature of human creativity that their drafters never had occasion to examine.

III. Legal Analysis

This Part examines the principal ownership claims advanced in respect of AI-generated fashion designs, beginning with the authorship problem that underlies them all.

A. The Authorship Vacuum and Its Doctrinal Consequences

The United States Copyright Office’s position, endorsed by the district court in Thaler v Perlmutter, is that copyright registration requires human authorship and that works produced autonomously by AI systems are not protectable.¹⁰ That position is constitutionally defensible, but analytically incomplete as applied to AI-assisted design. The Office’s guidance acknowledges that human-authored elements within an AI-generated work may be separately protectable, but offers very little on how to identify such elements when the human creative input operates entirely at the level of prompt engineering. The guidance creates a category without populating it. If AI-generated designs are categorically unprotectable, competitive advantage derives from secrecy rather than exclusivity—a shift that would drive AI-assisted fashion design toward trade secret protection and erode the public benefits that copyright’s disclosure-for-protection bargain is designed to produce.

B. Prompt Engineering as Creative Authorship

The most doctrinally plausible basis for copyright in AI-generated fashion designs lies in the creative investment of the human prompt-engineer, provided that engagement is genuinely substantive. Sophisticated commercial practice looks nothing like the caricature of a user who types “make me a blue dress” and accepts the first output. A designer working with a diffusion model may spend weeks refining hundreds of prompts, rejecting thousands of outputs, adjusting parameters, and iteratively steering the system toward a specific aesthetic vision shaped by accumulated knowledge of fashion history and material properties. The resulting design reflects creative choices about aesthetic register, historical reference, and cultural meaning that are properly attributed to the human designer even though the system executed the aesthetic realisation.

The analogy to photography is instructive, though its limits are as revealing as its explanatory force. In Burrow-Giles, the Supreme Court held that a photographer who had posed Oscar Wilde, selected and arranged costume and draperies, disposed the light and shade, and suggested the desired expression was an author.¹¹ An analogous argument holds for the fashion designer who carefully constructs iterative prompts embodying specific aesthetic decisions. The difficulty is one of structural degree rather than kind. Photography involved a one-step mechanical process in which the human’s creative choices were directly causally connected to the output. Generative AI involves a multi-step stochastic process mediated by a system of enormous complexity—one shaped by the creative labour of millions of uncredited human designers in the training dataset. That causal distance is not merely greater than in photography; it is differently structured in ways that matter for the originality inquiry, and the contribution-weighted framework must squarely confront it.

C. Developer Ownership Claims and the Work-for-Hire Analogy

AI developers have advanced, often implicitly through their terms of service, the argument that AI-generated outputs should vest in the developer as works for hire. The developer’s investment in building, training, and maintaining the system is substantial, and the system’s aesthetic capabilities reflect the developer’s creative and technical choices at the system level. The work-for-hire doctrine does not, however, translate cleanly. Under the Copyright Act 1976, a work made for hire is either prepared by an employee within the scope of employment or specially ordered in one of nine specified categories.¹² AI systems are not employees; their outputs are not commissioned in any legally cognisable sense; and fashion designs fall outside the enumerated categories. More fundamentally, vesting permanent copyright in AI developers would create concentrated monopoly rights in a small number of well-capitalised firms, effectively privatising the cumulative creative heritage of fashion at the precise moment AI makes that heritage most commercially accessible.

D. Fashion House Ownership: The Composite Claim

Fashion houses present a composite ownership claim: employer ownership of employee-generated work, licensee rights derived from AI developer terms of service, and independent design-right registrations for qualifying outputs. Section 11(2) of the Copyright, Designs and Patents Act 1988 vests copyright in an employer where a work is created by an employee in the course of employment,¹³ a rule that operates straightforwardly when a human employee uses an AI tool within their ordinary duties. The fashion house’s more ambitious claim—that it holds rights in AI-generated designs independently of any identified human author—runs into the same categorical barrier as the developer’s claim: corporate entities can hold copyright, but they cannot originate it. The EU design right framework is considerably more accommodating: registered Community design rights require novelty and individual character, not human authorship, entitling a fashion house that obtains registration for a qualifying AI-generated design to protection for up to twenty-five years. Design registration is faster, cheaper, and better calibrated to seasonal commercial cycles than copyright litigation—a point the fashion-law literature has not yet adequately absorbed.

E. Comparative Perspectives: United Kingdom, European Union, and India

Section 9(3) of the Copyright, Designs and Patents Act 1988 is unusual in the comparative landscape, supplying an author-substitute for computer-generated works by vesting ownership in “the person by whom the arrangements necessary for the creation of the work are undertaken.”¹⁴ What constitutes “arrangements” in the AI-generated fashion design context has received almost no judicial attention. Nova Productions Ltd v Mazooma Games Ltd confirmed that computer-generated elements may attract copyright under the provision, but it involved video game graphics, and its reasoning does not translate automatically to fashion design.¹⁵

Section 2(d)(vi) of India’s Copyright Act 1957 defines the author of a computer-generated work as “the person who causes the work to be created”—a formulation potentially encompassing both developers and sophisticated users without clear hierarchy between them.¹⁶ India’s Designs Act 2000 may in practice prove the most useful instrument,¹⁷ mirroring the EU experience and suggesting that the design-right route may be the de facto resolution of a copyright question that doctrine alone cannot resolve. Within the EU, the Commission’s 2020 Study acknowledged “legal uncertainty” surrounding AI-generated works but declined to recommend immediate legislative action—a restraint increasingly difficult to maintain as AI-generated fashion designs move from experimental to commercially routine.

IV. Case Law Discussion

A. Thaler v Perlmutter (D DC 2023)

Stephen Thaler sought to register copyright in a work produced by his “Creativity Machine” AI system, characterising it as the system’s autonomous output without human creative input. The Copyright Office refused registration, and the District Court affirmed in August 2023. Judge Howell’s opinion held that human authorship is “a bedrock requirement of copyright,” supported by the statute’s text, history, and the Constitution’s reference to “authors.”¹⁸ The court distinguished prior extensions of copyright to photography and film on the ground that in each instance a human being had made the creative choices that the copyright protected: the camera executes the photographer’s vision; it does not substitute its own.

Thaler confirms that wholly autonomous AI-generated works are categorically unprotectable in the United States—a conclusion with obvious implications for fashion houses that have represented AI-generated designs as entirely machine-produced. Beyond that proposition, however, the case is constrained by its own facts. Thaler’s claim rested on an explicit premise of zero human authorship; the opinion does not settle how much human creative input is sufficient to ground a copyright claim where human engagement operates at the level of prompt engineering rather than direct aesthetic execution. The deeper limitation of Thaler is epistemological: it proceeds as though the relevant choice is binary—purely human author or purely autonomous machine—when AI-assisted fashion design involves a continuous spectrum of human involvement. That binary frame provides inadequate guidance for the commercially significant middle ground, which is precisely where the doctrine needs to develop.

B. Feist Publications, Inc v Rural Telephone Service Co (1991)

Feist’s rejection of the “sweat of the brow” doctrine—the holding that copyright protects only the creative spark of original intellectual contribution, not mere investment or labour—has profound implications for AI-generated fashion design.¹⁹ A fashion house that invests heavily in building an AI design system cannot ground a copyright claim in that investment alone; it is the modern equivalent of the telephone company’s effort in compiling its subscriber database, and Feist is unambiguous that such effort is categorically insufficient. This is also why the developer ownership argument is doctrinally weak under existing US law. The AI developer’s investment—selecting architectures, curating datasets, adjusting hyperparameters—does not produce specific aesthetic features in any particular generated output; those features are produced by the model’s generative process applied to a specific user prompt. The causal and creative distance between the developer’s investment choices and the particular aesthetic qualities of any given output is precisely the gap that Feist holds the law will not bridge. That this disadvantages substantial investors is an argument for legislative supplement, not against Feist’s holding.

C. Naruto v Slater (9th Cir 2018)

In Naruto, the Ninth Circuit held that a crested macaque lacked statutory standing to sue for copyright infringement in a photograph it had taken by operating a photographer’s unattended camera. Critics have been too quick to dismiss its relevance to AI authorship debates. The court declined to extend copyright standing to a non-human animal despite its direct causal role in producing the photograph, on the ground that Congress had not expressly authorised animals to hold copyright and that the Act’s structure indicated a legislative focus on human creators.²⁰ The core reasoning translates to AI: statutory rights require statutory authorisation to extend to novel categories of claimants, and the existing categories do not include AI. Any framework for protecting AI-generated fashion designs in the United States must therefore be grounded in identifiable human authorship—not in the creative capabilities of the AI system, however impressive—making the quality and extent of human creative involvement the pivotal variable in every protection analysis.

V. Critical Analysis and Findings

A. The Failure of Existing Frameworks

Three systematic failures emerge from this analysis. The first is the categorical public-domain approach. Denying protection to a fashion designer who has spent weeks iteratively refining AI outputs through sophisticated prompt engineering is not a principled application of the human-authorship requirement but a doctrinal overcorrection that discounts real creative labour because it operates through a technological interface rather than a pencil or a loom. The question is not whether the designer used an AI; it is whether the designer’s use involved the kind of creative choice that copyright protects.

The second failure is the developer-ownership model. A legal regime that rewards the beneficiaries of unconsented mass appropriation of human creative work with permanent monopoly rights over the outputs generated from that appropriation is structurally perverse. The economic case for developer copyright is also weak under Feist: the developer’s investment is in the system, not in any particular design output, and the creative choices that produce aesthetic qualities in specific outputs are not the developer’s choices.

The third failure is the unreflective extension of work-for-hire analysis to fashion houses. Corporate ownership must be derivative of some human author’s original creative act. The fashion house’s claim to originate AI-generated design rights, independently of any identifiable human creative contribution, conflates economic organisation with creative authorship. The house may have excellent design-right and trade-secret claims; that is not the same as a copyright claim.

B. A Contribution-Weighted Entitlement Framework

The framework proposed here asks, at each stage of the AI design process, what substantive creative choices were made by identifiable human beings, and to what extent the aesthetic features of the AI-generated output reflect those choices rather than the autonomous generative capacity of the system. This is a structured inquiry with three sequential steps.

The first step charts human creative contributions across the design process: conceptualisation (the initial creative brief and aesthetic vision), prompt engineering (the iterative refinement of system inputs), curation (the selection of outputs for further development), and post-processing (human modification of AI outputs before commercial use). Each stage is assessed independently, since the creative quality of the contribution may vary substantially across them. The second step evaluates whether the identified contributions are original in the Feist sense—reflecting the contributor’s own intellectual creativity rather than routine professional choices any competent designer would make. The third step allocates ownership rights proportionally among the human contributors whose creative choices satisfy that threshold, subject to any applicable work-for-hire rules or contractual arrangements.

The framework holds distinct advantages over existing approaches. It preserves the constitutional and philosophical commitment to human authorship while accommodating the genuine creative investment that sophisticated AI-assisted design practice involves. It creates incentives for substantive human creative engagement rather than rewarding minimal prompt input. Courts familiar with contribution analysis from joint authorship and compilation cases will find it administrable within existing doctrine. And it provides a theoretically coherent basis for the partial protection that the Copyright Office’s guidance on AI-assisted works implicitly contemplates but has not yet adequately elaborated.

The objection that contribution-weighted analysis generates transaction costs the market cannot absorb has some merit, particularly in fast-fashion contexts where AI-generated design volumes are too high for routine case-by-case analysis. The appropriate response, however, is to develop standardised industry protocols—potentially administered through a collective licensing body modelled on ASCAP or the Copyright Licensing Agency—that allow contribution assessments to be made efficiently on an industry-wide basis. The music industry’s experience with collective licensing demonstrates that sophisticated rights management is achievable even in high-volume, multi-contributor creative environments.

VI. Conclusion

The question of who owns an AI-generated dress resists any simple answer. It does not follow, however, that the question is unanswerable, and the temptation to treat it as such—to wait for legislative resolution while commercial practice races ahead—should be resisted. The categorical denial of protection for AI outputs, the wholesale transfer of rights to AI developers, and the unreflective extension of work-for-hire rules to fashion houses each fail, for different reasons, to produce outcomes that are both doctrinally coherent and commercially workable. The contribution-weighted entitlement framework proposed here offers a more principled alternative: locate protection in measurable human creative investment at each stage of the AI design process, assess originality in accordance with established Feist standards, and allocate ownership to those whose contributions satisfy that threshold.

Courts can apply the framework within existing copyright doctrine across most jurisdictions without awaiting legislative revision. Legislatures—particularly in the United States, where the courts’ authority to innovate on authorship is constitutionally constrained and the Copyright Office’s administrative guidance cannot substitute for clear statutory text—will find in it a model for the statutory reform that is ultimately necessary. Policymakers face a more urgent charge: developing AI-specific intellectual property guidance before commercial practice outruns the law so thoroughly that retroactive correction becomes practically impossible. The fashion industry, meanwhile, should develop transparent protocols for documenting human creative contributions to AI-assisted design processes—documentation that will become increasingly valuable as ownership disputes, which have so far remained largely confidential and contractual, begin to reach public litigation.

Copyright law’s human-authorship requirement is not an administrative convenience or a historical accident. It expresses a considered judgment about the kind of creativity that intellectual property exists to protect and reward. That judgment need not be abandoned in the face of AI. What it requires is refinement: extension to new forms of human creative engagement with intelligent systems, precision about the varying degrees of creative contribution that AI-assisted practice involves, and clarity about the difference between protecting human creativity and protecting the economic interests of those who own the machines through which that creativity now flows. The dress may have been designed with artificial intelligence. The creative responsibility—for what it means, and for whom the law chooses to protect—remains irreducibly, and importantly, human.

Footnote(S):

¹ Star Athletica, LLC v Varsity Brands, Inc 580 US 405 (2017).

² Copyright, Designs and Patents Act 1988, s 9(3).

³ Burrow-Giles Lithographic Co v Sarony 111 US 53 (1884).

⁴ Berne Convention for the Protection of Literary and Artistic Works (1886, as revised) art 2.

⁵ Feist Publications, Inc v Rural Telephone Service Co 499 US 340 (1991).

⁶ Infopaq International A/S v Danske Dagblades Forening (C-5/08) EU:C:2009:465; Football Dataco Ltd v Yahoo! UK Ltd (C-604/10) EU:C:2012:115.

⁷ Star Athletica, LLC v Varsity Brands, Inc 580 US 405 (2017).

⁸ Council Regulation (EC) 6/2002 on Community designs [2002] OJ L3/1; Directive 98/71/EC on the legal protection of designs [1998] OJ L289/28.

⁹ Copyright, Designs and Patents Act 1988, s 9(3).

¹⁰ Thaler v Perlmutter No 1:22-cv-01564-BAH (D DC 18 August 2023).

¹¹ Burrow-Giles Lithographic Co v Sarony 111 US 53 (1884).

¹² Copyright Act 1976 (US), 17 USC § 101.

¹³ Copyright, Designs and Patents Act 1988, s 11(2).

¹⁴ Copyright, Designs and Patents Act 1988, s 9(3).

¹⁵ Nova Productions Ltd v Mazooma Games Ltd [2007] EWCA Civ 219.

¹⁶ Indian Copyright Act 1957, s 2(d)(vi).

¹⁷ Indian Design Act 2000.

¹⁸ Thaler v Perlmutter No 1:22-cv-01564-BAH (D DC 18 August 2023).

¹⁹ Feist Publications, Inc v Rural Telephone Service Co 499 US 340 (1991).

²⁰ Naruto v Slater 888 F 3d 418 (9th Cir 2018).

Bibliography

Table of Cases

• Acohs Pty Ltd v Ucorp Pty Ltd [2012] FCAFC 16
• Burrow-Giles Lithographic Co v Sarony 111 US 53 (1884)
• Feist Publications, Inc v Rural Telephone Service Co 499 US 340 (1991)
• Football Dataco Ltd v Yahoo! UK Ltd (C-604/10) EU:C:2012:115
• Infopaq International A/S v Danske Dagblades Forening (C-5/08) EU:C:2009:465
• Naruto v Slater 888 F 3d 418 (9th Cir 2018)
• Nova Productions Ltd v Mazooma Games Ltd [2007] EWCA Civ 219
• Star Athletica, LLC v Varsity Brands, Inc 580 US 405 (2017)
• Thaler v Perlmutter No 1:22-cv-01564-BAH (D DC 18 August 2023)

Table of Legislation

• Berne Convention for the Protection of Literary and Artistic Works (1886, as revised)
• Copyright Act 1976 (US), 17 USC §§ 101–1332
• Copyright, Designs and Patents Act 1988 (UK)
• Council Regulation (EC) 6/2002 on Community designs [2002] OJ L3/1
• Directive 98/71/EC on the legal protection of designs [1998] OJ L289/28
• Indian Copyright Act 1957
• Indian Design Act 2000
• Intellectual Property Act 2014 (UK)
• TRIPS Agreement, Annex 1C to the Marrakesh Agreement Establishing the World Trade Organization (1994)

Secondary Sources

• Andres Guadamuz, ‘Do Androids Dream of Electric Copyright? Comparative Analysis of Originality in Artificial Intelligence Generated Works’ [2017] 2 Intellectual Property Quarterly 169
• Estelle Derclaye (ed), Research Handbook on the Future of EU Copyright (Edward Elgar 2009)
• European Commission, Study on Copyright and New Technologies: Copyright Data Management and Artificial Intelligence (Publications Office of the European Union 2020)
• European Copyright Society, General Opinion on the EU Copyright Reform Package (2017)
• Jane C Ginsburg and Luke Ali Budiardjo, ‘Authors and Machines’ (2019) 34 Berkeley Technology Law Journal 343
• Jeanne C Fromer, ‘A Psychology of Intellectual Property’ (2010) 104 Northwestern Law Review 1441
• Lionel Bently and Brad Sherman, Intellectual Property Law (5th edn, Oxford University Press 2022)
• Madhavi Sunder, ‘IP3’ (2006) 59 Stanford Law Review 257
• Margot Kaminski, ‘The Capture of International Intellectual Property Law Through the US Trade Regime’ (2014) 87 Southern California Law Review 977
• Mark Bartholomew, ‘Protecting the Shape of Fashion: A Critique of the Star Athletica Decision’ (2018) 51 UC Davis Law Review 1891
• P Bernt Hugenholtz (ed), The Future of Copyright in a Digital Environment (Kluwer Law International 1996)
• Pamela Samuelson, ‘Allocating Ownership Rights in Computer-Generated Works’ (1986) 47 University of Pittsburgh Law Review 1185
• Paul Goldstein, Copyright: Principles, Law and Practice (3rd edn, Aspen Publishers 2005)
• Ryan Abbott, ‘I Think, Therefore I Invent: Creative Computers and the Future of Patent Law’ (2016) 57 Boston College Law Review 1079
• Shyamkrishna Balganesh, ‘Foreseeability and Copyright Incentives’ (2009) 122 Harvard Law Review 1569
• UK Intellectual Property Office, Artificial Intelligence and Intellectual Property: Copyright and Patents (2022)
• United States Copyright Office, Copyright and Artificial Intelligence, Part 1: Digital Replicas (2023)
• United States Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (2023) 88 Fed Reg 16190
• William F Patry, Patry on Copyright (Thomson Reuters, updated annually)
• World Intellectual Property Organization, Revised Issues Paper on Intellectual Property Policy and Artificial Intelligence (WIPO/IP/AI/2/GE/20/1 REV)