The source article (Elmahjub, 2025, Computer Law & Security Review 58:106170) argues from copyright's foundational justifications - utilitarian incentive, labor desert, and personality theory - that raw generative-AI outputs must remain in the public domain, while human creative contributions to AI-generated works may warrant protection, and it explicitly identifies as a critical gap in the literature the questions of how granular a user's intervention must be to count as authorship and how collective or layered authorship among model trainers, model owners, and end-users should be handled. The U.S. Copyright Office's Part 2 report on copyrightability (January 2025) holds that sufficiency of human contribution must be analyzed case-by-case and that prompts alone do not confer sufficient control over expressive elements; the D.C. Circuit's decision in Thaler v. Perlmutter (March 2025) settles only the wholly machine-generated end; and 2024-2026 scholarship continues to produce competing proposals without convergence, against persistent cross-jurisdictional divergence. The open problem is to formulate and justify operational legal criteria, usable by courts and copyright offices, for when human involvement in producing a generative-AI output suffices for the human contribution to qualify as independently copyrightable authorship - including how granular, iterative, or documentable that involvement must be across media - together with rules for attributing rights among multiple human contributors in layered production workflows.

Background

Copyright protects original works of authorship: works that a human created through at least a minimal degree of free and creative choice rather than by copying. Two doctrines anchor the present question. First, human authorship is a prerequisite: U.S. courts and the Copyright Office have long refused protection to works produced by a machine or mechanical process operating randomly or automatically without creative human input, and in Thaler v. Perlmutter the U.S. Court for the D.C. Circuit held in March 2025 that human authorship is required because many provisions of the Copyright Act make sense only if an author is a human being. Second, where a work mixes human and machine contributions, copyright protects the human author's original expression - their perceptible contributions, their creative selection, coordination, or arrangement of material, and their creative modifications - but not the machine-generated material itself; protection for such mixed works is often called 'thin' because it does not extend to the unmodified algorithmic portions.

Generative AI (GenAI) denotes large trained models - such as text-to-image diffusion models, which iteratively denoise random noise into images, and large language models - that produce text, images, audio, or video in response to user instructions called prompts. GenAI strains the above framework in a specific way: a prompt influences but does not determine the output, because the model samples probabilistically from statistical patterns absorbed during training on large corpora. Scholars therefore debate whether prompting is an act of authorship at all: Lemley (2024) argues that the model's probabilistic mediation of the prompter's intent complicates any direct attribution of the final expression to the human, while Craig and Kerr (2025) contend that the very notion of an 'AI author' is a category mistake because authorship is a relational, communicative human act. Official guidance has settled the extremes but not the middle: the U.S. Copyright Office's January 2025 report on copyrightability (Part 2 of its AI initiative) concluded that purely AI-generated material is unprotectable, that prompts alone do not currently provide sufficient human control over expressive elements, and that whether human contributions suffice to constitute authorship must be analyzed on a case-by-case basis - expressly declining bright-line rules. Jurisdictions diverge within that middle zone: the Beijing Internet Court in Li v. Liu (2023) protected a Stable Diffusion-generated image because the plaintiff's iterative prompting and parameter setting reflected his aesthetic choices and personal judgment, whereas other decisions and regimes, as surveyed in the source article, have drawn the line differently.

The source article (Elmahjub, 2025) supplies the doctrinal setting for the open question. It argues that the three classical justifications of copyright - the utilitarian incentive rationale, Lockean labor-desert theory, and Hegelian personality theory - all fail for raw GenAI outputs, which must therefore remain in the public domain, understood as the default baseline of unowned expressive material from which all creation draws. On this view a neighboring right or sui generis regime - a new, specially designed legal right outside classic copyright, of the kind that traditionally protects a producer's investment rather than an author's expression - is a possible legislative policy tool but not an interpretation of copyright law. Yet the article concedes the doctrinal knife-edge this leaves behind, stating that it highlights a critical gap in the literature: how granular must a user's intervention be to count as authorship, and how should collective or layered authorship scenarios be treated in which model trainers, model owners, and end-users each inject creative judgment at different stages of production? Under prevailing joint-authorship principles (built for collaboration among human authors, each of whom must supply independently copyrightable expression), mere ideas or general direction from one contributor do not suffice, which makes the allocation question in multi-actor AI workflows genuinely unsettled.

Problem Statement

Formulate and justify operational legal criteria - determinate enough for a copyright office examiner or a court to apply to a concrete work - that decide when a human's involvement in producing a generative-AI output suffices for the human contribution to qualify as independently copyrightable authorship. The criteria must specify what kinds and degree of human involvement count (prompting, iterative adjustment, selection, coordination, arrangement, and modification, singly or in combination) and how granular, repeated, or documentable that involvement must be, across the media to which the criteria are claimed to apply; they must cohere with existing originality and human-authorship doctrine, engaging rather than ignoring the decided authorities and official guidance that frame the gap - at minimum Thaler v. Perlmutter (D.C. Cir. 2025), the U.S. Copyright Office's January 2025 Part 2 copyrightability report, and the Beijing Internet Court's Li v. Liu judgment (2023); and they must include, or expressly justify the reduction of, an attribution rule allocating rights among multiple human contributors in layered production workflows (model trainers, model owners, and end-users, each injecting judgment at different stages). Adjacent policy questions that the source article acknowledges but does not pose as part of this doctrinal gap (mechanisms for remunerating upstream rights-holders whose works were used in training; the design of sui generis or neighboring rights for AI outputs; admissibility and confidentiality standards for workflow documentation; empirical measurement of the welfare effects of the public-domain baseline) are outside this problem's scope.

References

  1. Elmahjub, Ezieddin (2025). The algorithmic muse and the public domain: Why copyright's legal philosophy precludes protection for generative AI outputs. Computer Law & Security Review, vol. 58, article 106170. DOI: 10.1016/j.clsr.2025.106170; arXiv:2512.13750. https://arxiv.org/abs/2512.13750
  2. U.S. Copyright Office (2025). Copyright and Artificial Intelligence, Part 2: Copyrightability. Report of the Register of Copyrights, January 2025. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
  3. Thaler v. Perlmutter, No. 23-5233, 130 F.4th 1039 (U.S. Court of Appeals for the D.C. Circuit, decided March 18, 2025). https://www.copyright.gov/ai/docs/court-of-appeals-decision-affirming-refusal-of-registration.pdf
  4. Beijing Internet Court. Li v. Liu, (2023) Jing 0491 Min Chu No. 11279 (November 2023), English-language judgment. https://english.bjinternetcourt.gov.cn/pdf/BeijingInternetCourtCivilJudgment112792023.pdf
  5. Lemley, Mark (2024). How Generative AI Turns Copyright Law Upside Down. Science and Technology Law Review, vol. 25, no. 2. DOI: 10.52214/stlr.v25i2.12761. https://doi.org/10.52214/stlr.v25i2.12761
  6. Craig, Carys; Kerr, Ian (2025). The Death of the AI Author. In: Robot Law: Volume II, pp. 250-285. DOI: 10.4337/9781800887305.00014. https://doi.org/10.4337/9781800887305.00014
  7. Ramos-Zaga, Fernando A. (2025). Reconceptualizing Human Authorship in the Age of Generative AI: A Normative Framework for Copyright Thresholds. Laws, vol. 14, no. 6, article 84. DOI: 10.3390/laws14060084. https://doi.org/10.3390/laws14060084
  8. Zhang, Jinping; Guo, Danyang (2026). Copyright governance for GenAI-assisted works: Insights from Chinese judicial practice. Telecommunications Policy, vol. 50, article 103296. DOI: 10.1016/j.telpol.2026.103296. https://doi.org/10.1016/j.telpol.2026.103296