Own What You Create: Copyright Protection for AI-Assisted Works (Part II)
Part II – A Guide to Registering AI-Assisted Works
Copyright registration is a business decision with legal consequences. The good news for AI-assisted works, however, is that the registration process has remained largely the same and subject to the same principles. This article walks through what content creators and owners need to know to register AI-assisted works.
Why Registration Matters
Copyright protection arises automatically when a work is created and fixed in a tangible medium. Registration is not required to gain copyright protection, but it is critical to enforce your copyrights.
Registration (or preregistration) is generally a prerequisite to filing a copyright infringement action in a U.S. federal court for U.S. works. It is also a requirement for seeking statutory damages and attorney’s fees.[1] Specifically, a copyright owner who registers before infringement begins or within three months after first publication may seek statutory damages and attorney’s fees. A copyright owner who registers late, however, is generally limited to actual damages and profits. Registration also creates a public record of ownership and, if made within five years of first publication, establishes a presumption of validity.[2]
For creators and companies that generate valuable creative assets, AI-assisted or otherwise, registration should be treated as a standard step in the production process, not a reactive measure taken after a dispute arises.
The AI Disclosure Requirement
The Copyright Office requires disclosure of AI-generated content in a registration application when that content is more than “de minimis.”[3] This asks whether the AI-generated material is sufficiently creative to support a copyright claim if it had been created by a human author. If yes, it should be disclosed. If no, the AI contribution is “de minimis” and no disclosure is required. There is no quantifiable threshold that triggers the requirement; the standard is qualitative.
The following examples provided by the Copyright Office illustrate how the line is drawn in practice:
No disclosure required: An author uses AI to brainstorm plot ideas, character names, and chapter titles but writes the final text entirely herself. Because no AI-generated material appears in the final work, no disclosure is needed. Similarly, an author may use AI to check spelling and grammar, insert page numbers, and generate a table of contents. These contributions are de minimis because they would not be independently copyrightable if performed by a human. Therefore, no disclosure is required.
Disclosure required: An author writes the text of a graphic novel herself but uses AI to generate stylistic illustrations. The illustrations could be copyrightable if created by a human and should be disclosed. Likewise, a filmmaker uses AI to recreate background environments and generate special effects for scenes he independently shot. Those AI-assisted contributions are also appreciable and must be disclosed. In both situations, the AI-created elements must be disclaimed, leaving only the human-created elements for the Copyright Office’s review.
When disclosure is required, a brief and general statement is all the Office needs.[4] Applicants need not provide a detailed account of what AI tool was used or how; a short description in the “Limitation of Claim” screen will suffice — for example, “some text generated by AI” or “artwork generated by AI.”[5] Applicants can also provide a short description of the AI-generated materials in the “Note to Copyright Office” field on the Certification screen.
What Disclosure Means (and Doesn’t Mean)
In principle, the disclosure requirement is not a new obligation. The Copyright Office already requires applicants to identify unclaimable materials in an application, including previously published materials, public domain content, and materials owned by third parties. AI-generated content is treated as an additional category of unclaimable material.
The Copyright Office has also stated that disclosure of AI-created content “only occasionally results in refusal,” and that in most cases, a general statement identifying sufficient human authorship is enough to support a claim in copyright as to the human-authored portions.[6] The purpose of the disclosure is less about scrutinizing the application than it is about providing notice to the courts and the public that the registered work contains materials outside the copyright claim. As the Office has explained, the limitation of claim statement “simply put[s] courts and the public on notice that the work contains what the Office considers to be unclaimable material.”[7] The Office defers to courts to determine actual infringement and the scope of protection. It’s worth noting, however, that failure to disclose AI-generated content that the Office later discovers may result in the cancellation or narrowing of a previously granted registration. For example, after discovering a creator’s social media posts disclosing their use of Midjourney to create images contained in an already-registered graphic novel, the Copyright Office cancelled the prior registration and replaced it with a narrowed registration that excluded the AI-generated images.[8]
One practical point: AI-generated material that is disclosed and excluded from the claim does not need to be removed from the deposit. The complete work—including both human-authored and AI-generated elements—should be submitted as is.[9] The registration will cover only the human-authored portions identified in the application.
Registration Timing: Strategy and Best Practices
Ensuring timely registration of your copyrights is important for two reasons. First, it can affect the scope and clarity of the copyright claim. Second, it preserves important remedies in an infringement action.
On scope, content owners and creators may benefit from registering the human-authored version of a work before making any modifications using AI. Doing so preserves a clear copyright claim to the underlying human-authored work, even if AI is later used to make appreciable changes, and even if the AI-modified version of the work is what is ultimately distributed. The same principle may apply to derivative works. If AI is used to generate a translation, adaptation, or stylistic variation of a human-authored work, registration of the work may still help address unauthorized uses of the later version. That said, registration of the original work does not itself extend protection to later-added AI-generated material, and the extent to which this approach will hold up in court remains to be seen.
On remedies, a copyright owner who registers within three months of first publication preserves eligibility, in many cases, to elect statutory damages and attorney’s fees in an infringement action.[10] A copyright owner who registers late is limited to actual damages and the infringer’s profits, which may be difficult to prove and may not justify the cost of federal litigation. This distinction matters more than it may appear. The availability of statutory damages is often what makes an infringement claim economically viable; without it, many valid claims go unenforced simply because the cost of litigation exceeds what a plaintiff can realistically recover. Registration of core creative assets within the three-month window should therefore be treated as a standard production step, not a reactive measure taken after a dispute arises.
Conclusion
Registering AI-assisted works is less complicated than it may appear. The Copyright Office has tried to make the process accessible, and the disclosure requirement, while new in its application to AI, follows principles that have governed registration for decades. For most mixed human-AI works, the result of proper registration is a valid copyright in the human-authored portions, a clear public record of what is and isn't claimed, and the ability to enforce that claim in federal court.
The more important decisions are strategic ones: what to register, when to register it, and how to structure the workflow so that human authorship established is well documented before AI enters the picture. Those decisions, made early and consistently, will determine whether a copyright registration is a meaningful asset or a limited one.
Part III, to follow, will address best practices when considering contracting and licensing around AI-assisted works.
Registration: Key Action Items
☐ Register early. Where possible, register within three months of publication to preserve important remedies.
☐ Choose the right version to register. Consider whether the clearest claim lies in the original human-authored work, a later mixed human-AI version, or both.
☐ Claim only the human-authored portions. Identify what was created by a human and exclude AI-generated material from the claim.
☐ Align the application with the workflow record. Keep drafts, source files, prompts, outputs, and editing history so the registration can be supported if later challenged, as previewed in Part I of this series.
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]17 U.S.C. § 412; 17 U.S.C. § 504(c).
[2]17 U.S.C. § 411(a).
[3]Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190, 16,193 (Mar. 16, 2023) (“Registration Guidance”).
[4] Id.
[5] Copyright Office Webinar, Application Process for Registration of Works with Artificial Intelligence-Generated Content, Transcript at 2 (June 28, 2023) (“Webinar Transcript”).
[6] Id.
[7] Id. at 8.
[8] Letter from Robert J. Kasunic, Assoc. Register of Copyrights, to Van Lindberg, Re: Zarya of the Dawn (Registration # VAu001480196) (Feb. 21, 2023), available at https://www.copyright.gov/docs/zarya-of-the-dawn.pdf.
[9] Id.
[10] 17 U.S.C. § 412; 17 U.S.C. § 504(c).