Copyright protects only works authored by humans, but generative artificial intelligence now produces content through intertwined human prompting, selection, editing, and curation, leaving registries and courts without a determinate rule for when the human contribution suffices. The open question, posed by Chesterman (2024) in 'Good models borrow, great models steal: intellectual property rights and generative AI', is what level of human involvement — input or control — is required for an AI-assisted work to count as authored by a human, a threshold whose determination the paper notes 'is left to the courts,' and by what principled, administrable criterion should courts and legislatures decide it?

Background

Copyright is the legal regime that grants creators exclusive rights over original works of authorship such as texts, images, music, and code, but it protects only works that originate from a human author. In the United States, the Supreme Court held in Burrow-Giles Lithographic Co. v. Sarony (1884) that an 'author' is the human 'to whom the work owes its origin,' and modern doctrine requires only independent creation plus a 'modicum of creativity' (Feist Publications v. Rural Telephone Service, 1991); the U.S. Copyright Office (USCO), the agency that administers registration, has required since 1973 that a work owe its origin to a human agent. Generative artificial intelligence (AI) — systems such as large language models (LLMs, neural networks trained on text that generate prose in response to instructions) and text-to-image diffusion models (which generate images from text prompts) — unsettles this framework because creative agency is now distributed between a human user and a machine. Human involvement spans a spectrum: issuing a single natural-language instruction (a prompt); iterating through many prompts and refinements; selecting among generated candidates (curation); directly editing or modifying machine output; arranging AI-generated material within a larger human-authored work; and using AI as an ordinary tool inside an otherwise human creative process. In a series of registration decisions — notably the graphic novel 'Zarya of the Dawn' and the image 'Théâtre D'Opéra Spatial' (both 2023) — the USCO protected the human contributions (text, selection, arrangement) while refusing protection to the AI-generated elements, and in March 2023 it issued guidance stating that copyrightability of mixed works is a case-by-case inquiry in which prompts, with then-current technology, 'function more like instructions to a commissioned artist' and do not give users ultimate creative control. Other jurisdictions diverge: the United Kingdom has had since 1988 a statutory category for 'computer-generated' works whose author is the person by whom the arrangements necessary for creation are undertaken (retained after a 2021–2023 government review); the Beijing Internet Court in Li v. Liu (November 2023) held an image generated with the Stable Diffusion text-to-image model protectable because the user's roughly 150 prompts plus subsequent adjustments reflected personalized intellectual investment; Japan's May 2024 governmental guidelines list case-by-case factors such as the amount of instructions and the number of generation attempts; and the European Union assesses originality through the Court of Justice of the European Union (CJEU) standard that a work be the author's 'own intellectual creation.' Samuelson (2023) documents how concurrent litigation over AI training data sharpened attention on both sides of the problem — inputs and outputs. Chesterman (2024) posed the threshold itself as an open question: discussing the U.S. Copyright Office Compendium's refusal to register works produced 'without any creative input or intervention from a human author,' it observed that the word 'any' 'begs the question of what level of human involvement is required to assert authorship,' and, discussing Singapore's Copyright Act 2021, that 'AI-assisted works may still warrant protection if there is a causal connection to a human exercising input or control, though determining the threshold for that connection is left to the courts' — an indeterminacy that persists especially where prompting, editing, curation, and machine generation are intertwined. Subsequent developments leave the threshold undecided: the U.S. Copyright Office's January 2025 report 'Copyright and Artificial Intelligence, Part 2: Copyrightability' holds that the sufficiency of human contribution must be analyzed case-by-case and that prompts alone, on current technology, do not provide sufficient control, while the D.C. Circuit's 2025 decision in Thaler v. Perlmutter affirms the human-authorship requirement for wholly machine-generated works without fixing the boundary for assisted works; the structured authorship-threshold tests and information-theoretic measures of human contribution proposed through 2026 remain unadopted.

Problem Statement

For copyright regimes built on the human-authorship and originality requirement — such as the U.S. registration standard refusing protection to works 'produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author' and the EU 'own intellectual creation' standard — determine the threshold of human creative input or control required for a work produced with generative AI to qualify as authored by a human, and therefore eligible for copyright protection, and articulate a principled, administrable doctrinal standard, statutory rule, or evaluative framework by which courts, copyright registries, and legislatures can apply that threshold consistently. Regimes that do not require a human author for computer-generated works (the United Kingdom's CDPA 1988 category) or that protect AI-generated output on other grounds (China's Beijing Internet Court approach) serve as comparative context; an answer may be regime-specific or comparative and is not required to produce a single standard simultaneously binding all divergent regimes. The question, posed by Chesterman (2024), who asked 'what level of human involvement is required to assert authorship' and noted that 'determining the threshold for that connection is left to the courts,' spans the full spectrum of intertwined human-machine contribution: single and iterative prompting, selection and curation among generated candidates, editing and modification of machine output, arrangement of AI-generated material within larger human works, and tool-like uses embedded in human creative processes. The criterion must locate the authorship line across this spectrum, be grounded in existing authorship and originality doctrine so that adjudicators can apply it in concrete cases, and specify the conditions under which human contribution suffices rather than merely restating that sufficiency is assessed on a case-by-case basis.

References

  1. Simon Chesterman (2024), 'Good models borrow, great models steal: intellectual property rights and generative AI', Policy and Society 44(1):23-37, DOI: 10.1093/polsoc/puae006, https://doi.org/10.1093/polsoc/puae006 (source paper; submitted draft: NUS Law Working Paper No 2023/025, https://law.nus.edu.sg/wp-content/uploads/2023/10/025_SimonChesterman.pdf)
  2. U.S. Copyright Office (2023), 'Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence', 88 Fed. Reg. 16190 (March 16, 2023), https://www.federalregister.gov/documents/2023-03-16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence, permanent copy at https://www.govinfo.gov/content/pkg/FR-2023-03-16/html/2023-05321.htm
  3. 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
  4. Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), No. 23-5233, decided March 18, 2025, opinion available at https://media.cadc.uscourts.gov/opinions/docs/2025-03-23-5233.pdf, mirror at https://www.courtlistener.com/opinion/10357586/stephen-thaler-v-shira-perlmutter/
  5. Tingting Wen (2024), 'Beijing Internet Court recognizes copyright in AI-generated image', Journal of Intellectual Property Law & Practice 19(3):250-255, DOI: 10.1093/jiplp/jpad127, https://doi.org/10.1093/jiplp/jpad127
  6. Pamela Samuelson (2023), 'Generative AI meets copyright', Science 381(6654):158-161, DOI: 10.1126/science.adi0656, https://doi.org/10.1126/science.adi0656
  7. Julija Kalpokienė (2025), 'Generative artificial intelligence - is there still room for a human author?', ERA Forum 26(4):561-575, DOI: 10.1007/s12027-025-00856-6, https://doi.org/10.1007/s12027-025-00856-6
  8. Nirogini Thambaiya, Kanchana Kariyawasam, Chamila Talagala (2025), 'Copyright law in the age of AI: analysing the AI-generated works and copyright challenges in Australia', International Review of Law, Computers & Technology 39(3):448-473, DOI: 10.1080/13600869.2025.2486893, https://doi.org/10.1080/13600869.2025.2486893
  9. Zihang Lan (2026), 'GenAI as Fictional Author: Statutory Attribution of Copyright in AIGC', International Journal of Digital Law and Governance, DOI: 10.1515/ijdlg-2026-0008, https://doi.org/10.1515/ijdlg-2026-0008
  10. Yueqi Xie, Tao Qi, Jingwei Yi, Xiyuan Yang, Ryan Whalen, Junming Huang, Qian Ding, Yu Xie, Xing Xie, Fangzhao Wu (2024), 'Measuring Human Contribution in AI-Assisted Content Generation', arXiv:2408.14792, https://arxiv.org/abs/2408.14792
  11. Ulaş Göker Bektaş (2026), 'The Algorithmic Authorship Threshold: identifying human contribution in AI-assisted creativity', Journal of Intellectual Property Law & Practice 21(9):610-618, DOI: 10.1093/jiplp/jpag078, https://doi.org/10.1093/jiplp/jpag078
  12. Geyu Deng (2026), 'Human Authorship in AI-Assisted Creative Works: A Three-Factor Test for Evaluating Expressive Control', International Journal of Law and Politics Studies 8(5):01-07, DOI: 10.32996/ijlps.2026.8.5.1, https://doi.org/10.32996/ijlps.2026.8.5.1
  13. Lan Li, Noelle Li Ying Cheah, Seung Hyun Kim (2025), 'AI art in the gig economy: Investigating the effects of non-copyrightability in online labor markets', Decision Support Systems 199:114545, DOI: 10.1016/j.dss.2025.114545, https://doi.org/10.1016/j.dss.2025.114545