Own What You Create: Copyright Protection for AI-Assisted Works (Part I)
Part I – The Human Authorship Requirement and the Scope of Protection
AI is now standard equipment for the modern workforce. Designers use it to generate visual concepts. Writers use it to outline drafts. Studios use it to generate special effects. The tools vary, but one common question prevails: what copyrights, if any, can you claim over the output?
The answer turns on the degree of human control over the creative expression contained in the work. Under current U.S. copyright law, there is no protection for expression generated solely by AI. There is also no simple formula for how much human involvement is enough to ensure copyrightability. This article, the first in a series addressing copyright protection for AI-assisted works, examines the legal framework governing that question and the practical considerations for content owners and creators who use AI in their workflow.
The Human Authorship Requirement
Copyright law has always contemplated a human author. Writing for the Supreme Court in 1903, Justice Holmes observed that “[p]ersonality always contains something unique”—that even a modest work of art contains “something irreducible, which is one man’s alone.”[1] Courts have applied the same principle to refuse copyright protection for works produced by animals,[2] and, more recently, by machines.
Developments in generative AI have cemented the human authorship requirement in copyright law. The Copyright Office issued registration guidance in 2023 confirming that it will not register works produced by machines without creative input.[3] In March 2025, the D.C. Circuit affirmed in Thaler v. Perlmutter that the Copyright Act requires all eligible work to be “authored in the first instance by a human being.”[4] In March 2026, the United States Supreme Court denied certiorari, leaving the D.C. Circuit decision to stand. The question that remains unsettled, however, is how much human involvement is sufficient to support a copyright claim in works where AI and human creativity are both present.
How Much Human Authorship Is Enough?
The Copyright Office’s guidance provides that copyright protection turns on whether a human author controlled the expressive elements of an AI-assisted work.[5] In other words, was AI merely used to assist the creative process, or did AI generate the creative elements subject to human direction? The same tool, used differently, produces a different legal result.
A multi-tier analysis illustrates the range of protection, from broadest to most limited:
Tier 1 – AI as process tool only. Copyright protection will be the broadest when the range of creative expressions was entirely man-made, and AI is used only for tasks that do not affect the expressive content of the work. Examples: a novel written independently by an author who uses AI only to brainstorm plot ideas; a design sketched by a designer who uses AI to conduct market research; a melody composed and performed by a musician who uses AI only to convert the recording into written sheet music. In each of these examples, the creative output is entirely attributable to a human being.
Tier 2 – AI as tool to refine human expression. Where AI is used only to refine or execute creative choices that a human author has already made, copyright protection likely remains strong. For example, an animator designs a character from original sketches, including specific features, proportions, expression, color, then uses AI to smooth animation, interpolate frames, or apply consistent styling. Here, the AI has no role in creative decision making; its only role is to execute certain “refinements” that are likely not protectable on their own. Because the expressive content (character design) is manmade, the scope of copyright likely remains broad.
The critical requirement to ensure copy protection is documentation. Creators and companies should preserve evidence of the workflow and sequencing: what the human authored first, what was subsequently submitted to AI, and what AI contributed. That record is the foundation of the copyright claim if it is ever challenged.
Tier 3 – Human-directed, AI-generated expression. Where AI is used to generate creative expressions at human instruction, the scope of protection becomes more limited. The Copyright Office’s decision regarding Zarya of the Dawn is instructive. There, an artist used Midjourney to generate images for a graphic novel, selected and arranged the images, and added text. The Copyright Office approved protection for the text and the selection and arrangement, but denied protection for the AI-generated images on the basis that it was Midjourney—not the artist—that made the specific visual choices in translating prompts to images.[6] The Office has confirmed this analysis in subsequent guidance, noting that prompts do not give users sufficient control over expressive elements because AI systems are unpredictable, and the same prompt can generate varying outputs.[7] [8]
The key takeaway is that where expressive elements are generated by AI but edited and compiled by humans, a compilation copyright may be available for the overall selection and arrangement of the work, even if no individual AI-generated elements are protectable.[9] For a compilation copyright to apply, creators will need to show that the human-made curatorial choices for the selection and arrangement are original enough to warrant protection.[10] Documentation is again critical here to protect against challenges to copyrightability. Creators should preserve a record of how human decisions were made in the process of modifying AI output and making selections and arrangements, i.e. which images or elements were chosen and why, what edits were applied, and how the overall work was assembled. That record distinguishes protectable curatorial judgment from mere acceptance of AI-generated results.
Practical Implications for Content Creators and Owners
The framework above has direct implications for how creative workflows are documented. The practical guidance that follows is organized by sector, but the best practices around workflow documentation apply equally across all industries to safeguard against potential challenges to copyrightability.
Individual creators. The most important practical step is to build documentation into the creative process from the start. Save dated drafts, sketches, and source files that show what you created before AI entered the workflow. If you use AI to refine or edit your work, keep a record of what you submitted to AI (documents, prompts, etc.) and what the AI output was. If you use AI to generate elements you then select and arrange, document your selection criteria and editorial choices, and have a record of the different iterations and versions of the work. A contemporaneous record of your creative process is more credible than a reconstruction prepared after a dispute arises.
Media and entertainment companies. Companies that own portfolios of creative properties—animated franchises, character IP, show libraries—should pay close attention to chain of title. The value of media assets depends on the scope and defensibility of their copyrights, and AI use in production can complicate both. Character art, background environments, and visual assets generated through AI workflows without adequate documentation may create gaps in the copyright foundation of core IP that surface later in licensing, enforcement, or transaction contexts. Legacy works with a clean copyright history and newer AI-assisted productions will present different issues and warrant different analyses. For new productions, the practical priority is the same as for individual creators: document what was created by humans, what was submitted to AI, and what AI generated. For existing properties, a chain-of-title review is recommended if AI-assisted assets have been incorporated into registered works.
Video game and mobile game developers. Game studios may use AI across multiple stages of production—art, audio, UI, narrative—and the scope of protection may be different for each. Likewise, game engines, scripts, and gameplay logic generated through AI coding tools may carry the same copyright uncertainty as AI-generated visual assets, further complicating the copyright analysis. But even where individual AI-generated assets lack copyright protection, the human creative judgment reflected in how those elements are selected, arranged, and designed could support a compilation copyright for the game as a whole. The practical priority is to treat design documentation as a legal asset: record the human decisions behind level design, UI logic, visual hierarchy, and asset selection. Studios should also establish internal workflows that require human review and sign-off at key creative decision points and a contemporaneous record of those decisions.
What Comes Next
Human authorship is well-settled in copyright law. The open question is how much human involvement through prompting and iteration, if any, is sufficient to support a copyright claim in works that combine human and AI-generated expression. Allen v. Perlmutter, pending in the District of Colorado, will likely be the first court ruling on that question. There, Jason Allen used more than 600 prompts across hundreds of iterations in Midjourney to generate an image that he modified in Photoshop and Gigapixel AI. The Copyright Office denied registration on the grounds that Allen lacked control over how Midjourney responded to his prompts.[11] A ruling is expected in the first half of 2026.
Part II of this series will address guidance and best practices around registering AI-assisted works as a means to protect creative assets.
Creative Development: Key Action Items
☐ Document human creative contributions from the start. Save dated drafts, sketches, and source files that show what was created before AI entered the workflow.
☐ Track what was submitted to AI and what AI generated. Keep a record of prompts, inputs, and outputs at each stage of the creative process.
☐ Document selection and arrangement decisions. If AI-generated elements are selected and compiled, preserve a record of which elements were chosen, why, by whom, and how the work was assembled.
☐ Treat design documentation as a legal asset. For companies, require human review and sign-off at key creative decision points and maintain a contemporaneous record of those decisions.
Chieh Tung is a litigator representing companies in copyright, trademark, and business disputes. She writes about developments at the intersection of AI and intellectual property.
***
[1] Bleistein v. Donaldson Lithographing Co., 188 U.S. 239, 250 (1903).
[2] Naruto v. Slater, 888 F.3d 418, 426 (9th Cir. 2018).
[3] Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190, 16,192 (Mar. 16, 2023) (“Registration Guidance”) at 16,192.
[4] Thaler v. Perlmutter, 130 F.4th 1039, 1044 (D.C. Cir. 2025).
[5] U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability 7–8 (Jan. 29, 2025) (“Part 2 Report”); Registration Guidance, supra note 4, at 16,192.
[6] Registration Decision on Zarya of the Dawn, Copyright Office Letter to Kristina Kashtanova (Feb. 21, 2023), available at https://www.copyright.gov/docs/zarya-of-the-dawn.pdf.
[7] Id.
[8] Part 2 Report, supra note 5, at 17–18.
[9] Zarya of the Dawn Registration Decision, supra note 6.
[10] Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340, 348 (1991); Part 2 Report, supra note 5, at 22–23.
[11] Copyright Review Board Decision on Théâtre D’Opéra Spatial, Copyright Board Proceeding No. 1-11743923581 (Sept. 5, 2023); Allen v. Perlmutter, No. 1:24-cv-02665 (D. Colo., filed Sept. 13, 2024).