The Human Authorship Requirement for AIGenerated Works: A Comparative Analysis of US UKand EU Copyright Law

1. Introduction

Generative artificial intelligence (AI) tools such as ChatGPT, DALL-E and Midjourney now produce text, images and other creative outputs at scale, raising fundamental questions for copyright doctrine built around human creators. These systems can generate content without direct human oversight at the moment of fixation, challenging long-standing assumptions that authors are natural persons exercising intellectual judgement. Across jurisdictions, courts and policy-makers have had to decide whether AI-generated outputs qualify as “works” and, if not, how to treat human–AI collaborations and the economic interests at stake.

This article examines the requirement of human authorship in US, UK and EU copyright law and evaluates whether existing frameworks are conceptually and normatively adequate for AI-generated works. It first outlines the doctrinal foundations of the human authorship requirement and its link to originality. It then considers key case law and administrative practice, especially in the United States, before turning to the EU’s “author’s own intellectual creation” standard and the UK’s sui generis provision for computer-generated works. Finally, it draws on recent scholarship to assess reform options, including collaborative authorship models, sui generis rights and public-domain approaches.

2. Doctrinal Foundations: Human Authorship and Originality

Copyright systems have historically presupposed that authors are human beings. The Berne Convention’s structure and history indicate that “author” refers to a natural person whose intellectual labour gives rise to a protected work. National doctrines then tie protection to originality, understood as a threshold of intellectual creativity attributable to that person. In US law, originality requires independent creation and a minimal degree of creativity, articulated famously in Feist as the “sine qua non of copyright” and “bedrock requirement”.

Li explains that the originality standard is conceptually intertwined with human authorship, because originality presupposes creative decision-making by a conscious agent rather than mechanical reproduction. In EU law, the Court of Justice requires that a work be the “author’s own intellectual creation”, reflecting “free and creative choices” that express the author’s personality. This personality-based test strongly implies an anthropocentric understanding of authorship that AI systems, lacking legal and moral personality, cannot satisfy.

These doctrinal choices are not purely semantic. They implement several policy objectives: incentivising human creativity, avoiding perpetual rights for non-human entities, and preserving the link between moral rights and personal authorship. As Mammen and colleagues argue, both copyright and patent law treat human creativity as normatively special, not merely because human outputs can be novel, but because creative acts occur within subjective and social processes that current AI systems do not share.

3. United States: Human Input and Administrative Practice

US law contains no explicit statutory definition of “author”, but the Constitution’s Intellectual Property Clause and the Copyright Act are drafted on the implicit assumption that authors are persons. The US Copyright Office’s Compendium crystallises this by stating that works must be “created by a human being” and expressly refusing registration for works produced by nature, animals or processes operating “without any creative input or intervention from a human author”.

Case law confirms this anthropocentric reading. In Burrow-Giles the Supreme Court linked “author” to a human who produces an “original intellectual conception”. In Naruto v Slater, the Ninth Circuit rejected copyright claims brought on behalf of a macaque, holding that non-human animals lack standing under the Copyright Act and noting the Office’s refusal to protect animal-created works. More recently, in Thaler v Perlmutter, the District Court for D.C. upheld the Office’s refusal to register a work listed as authored solely by an AI system, describing human authorship as a “bedrock requirement” that courts could not set aside without legislative change.

The Office’s 2023 policy statement on AI-generated works introduces a finer-grained approach to human–AI collaborations. It distinguishes between:

• Outputs generated autonomously by AI, which are categorically excluded from protection; and
• Works where humans use AI as a tool but exercise “sufficient creative control”, for example by selecting, arranging and editing AI outputs, which may attract protection for the human-authored elements only.

Li analyses this emerging practice through the lens of human contribution: protection hinges on whether the human contribution satisfies the minimal creativity threshold independently of the machine’s statistical processes. Halwachi similarly argues that existing US originality doctrine can accommodate AI-assisted works when humans retain meaningful input, but struggles with works where AI is the primary creative agent.

4. EU and UK: Author’s Own Intellectual Creation and Computer-Generated Works

The EU’s copyright acquis lacks a horizontal definition of “author”, yet multiple directives and CJEU decisions make clear that authorship is anthropocentric. For example, the software and database directives refer to authors as “natural persons” or groups thereof, while allowing legal persons to be designated right holders under national law. In Painer, the Advocate General expressly stated that “only human creations are therefore protected” and the Court treated a portrait photograph as protectable where it reflected the photographer’s intellectual creation.

The CJEU’s case law on originality further reinforces this. In Infopaq and Football Dataco, the Court held that a work or database is protected only where structure or selection is not dictated solely by “technical considerations, rules or constraints”, but instead leaves room for creative freedom. Halwachi notes that, transposed to AI, this criterion suggests difficulties for outputs largely determined by algorithmic training data and model architecture rather than by a human author’s free choices. Where AI systems generate content with minimal human input, it becomes hard to identify the “author’s own intellectual creation” in the EU sense.

The UK occupies a somewhat distinctive position. While adhering to the Berne framework and the EU-shaped originality standard, it also retains a sui generis provision for “computer-generated works” in the Copyright, Designs and Patents Act 1988, which designates as author “the person by whom the arrangements necessary for the creation of the work are undertaken”. Li and Halwachi both highlight this as a possible foothold for protecting certain AI-generated works, because it explicitly contemplates works where “there is no human author”.

However, the provision predates modern generative AI and its scope remains uncertain. It is unclear how minimal “arrangements” can be – whether merely prompting an AI system suffices, or whether more substantial design, training or curation is required. Moreover, the UK’s originality requirement still demands a “modicum of creativity”, raising questions about whether mere technical configuration of a model, absent creative decisions about individual outputs, should attract authorship.

5. Comparative Perspectives and Emerging Scholarship

Taken together, these sources reveal a strong cross-jurisdictional convergence on human authorship as a prerequisite for copyright, albeit grounded in different rationales. Halwachi frames this as a “dilemma of authorship”, arguing that both US and EU systems treat authorship as a human trait, yet neither has enacted legislation specifically tailored to AI-generated works. Li shows that courts in Asia have occasionally experimented with more flexible approaches, including a Chinese decision assigning rights in an AI-generated news article to the system’s developer, but emphasises that such rulings remain isolated and contested.

Mammen and colleagues add a multidisciplinary dimension by unpacking creativity into external, subjective and social components. External creativity concerns the novelty and value of the artefact; subjective creativity tracks the psychological processes of problem-selection, mind-wandering and evaluative judgement; and social context influences what counts as valuable and how creators develop their ideas. They argue that IP law implicitly values not just the artefact’s external qualities but also the fact that it emerges from human cognitive and social processes, which current AI systems do not possess. On this view, even highly sophisticated AI outputs that resemble human works externally may still fail to meet the deeper creativity rationale underlying copyright.

Both Li and Halwachi survey reform proposals that respond to this tension. Collaborative models treat AI-assisted works as joint creations, with authorship allocated to humans who contribute prompts, curation and post-editing, provided their contribution reaches the originality threshold. Sui generis rights would grant limited protection to AI outputs based on investment rather than creativity, echoing the EU database right. Public-domain approaches deny any exclusive rights in purely AI-generated works, reasoning that the absence of human authorship should leave such content freely reusable and that AI developers can rely on trade secrets, service models and first-mover advantages for incentives.

6. Evaluation and Future Directions

From a doctrinal standpoint, the current insistence on human authorship provides coherence with long-standing principles of originality and moral rights, and avoids radical re-engineering of copyright systems to accommodate non-human creators. It also preserves international harmonisation anchored in Berne and TRIPS, which assume human authorship. However, as Li and Halwachi indicate, the rapid growth of generative AI exposes practical difficulties: courts face hard borderline cases where human and machine contributions are intertwined, and stakeholders confront legal uncertainty about the status of AI-heavy works.

Normatively, Mammen’s enriched account of creativity suggests caution about equating AI “outputs” with human works merely because they appear similar. If copyright aims to recognise and foster human creative agency embedded in social contexts, extending full protection to autonomous AI outputs risks diluting that purpose. At the same time, denying any protection for human–AI collaborations where human choices genuinely shape the outcome could chill legitimate creative experimentation with new tools.

A plausible medium-term path is incremental refinement rather than wholesale reform. That might include clearer administrative guidance on how much human control and editing is required for protection; modest statutory clarification of concepts such as “computer-generated works” in the UK; and, potentially, limited sui generis regimes for certain AI-generated outputs, carefully designed to avoid over-protection. In all scenarios, transparency about AI involvement in the creative process and careful documentation of human contributions will be crucial for both rights claims and user certainty.

Bibliography 

Gaidartzi A and Stamatoudi I, ‘Authorship and Ownership Issues Raised by AI-Generated Works: A Comparative Analysis’ (2025) 14 Laws 57.

Li Y, ‘The Human Authorship Requirement in AI-Generated Works: A Comparative Analysis of Copyright Protection Frameworks’ (2025) Journal of Law and Governance 21.

US Copyright Office, Compendium of US Copyright Office Practices (3rd edn, 2021)​

US Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (Federal Register, 30 August 2023).

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).

Painer v Standard VerlagsGmbH (C-145/10) EU:C:2011:798.

Stephen Thaler v Shira Perlmutter No 22-1564 (D DC, 18 August 2023).

Zarya of the Dawn (US Copyright Office refusal letter, 21 February 2023).

Copyright, Designs and Patents Act 1988 (UK)

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 OJ L167/10 (InfoSoc Directive) art 2(a).​

Directive 2009/24/EC of the European Parliament and of the Council of 23 April 2009 on the legal protection of computer programs OJ L111/16 art 2(1).

Directive 96/9/EC of the European Parliament and of the Council of 11 March 1996 on the legal protection of databases OJ L77/20 art 4(1)