Copyright Protection for AI-Generated Content Under U.S. Law

U.S. copyright doctrine is under sustained pressure from a single structural question: whether output produced by an artificial intelligence system qualifies for federal copyright protection, and if so, who holds that protection. The Copyright Act of 1976 and decades of judicial interpretation anchor protection to human authorship, creating a framework that did not anticipate autonomous machine creation. This page examines the statutory text, administrative guidance from the U.S. Copyright Office, key adjudicated cases, and the classification disputes that arise when human and machine contributions combine in a single work.

• Definition and Scope
• Core Mechanics or Structure
• Causal Relationships or Drivers
• Classification Boundaries
• Tradeoffs and Tensions
• Common Misconceptions
• Checklist or Steps
• Reference Table or Matrix
• References

Definition and Scope

Copyright protection under U.S. law attaches to "original works of authorship fixed in any tangible medium of expression" (17 U.S.C. § 102(a)). The statute does not define "author" with precision, but the U.S. Copyright Office and federal courts have consistently interpreted the word to require a human creator. Works that lack a human author — including those autonomously generated by software or AI systems — fall into the public domain upon creation and receive no federal copyright protection.

The scope of this doctrine extends to text, images, music, code, and any other expressive output. The operative legal boundary is not whether a computer was used in creation, but whether a human exercised creative control over the protectable expression. That distinction shapes every downstream question about registration, ownership, licensing, and infringement — topics that intersect substantially with the broader treatment of AI and intellectual property law.

Core Mechanics or Structure

The Human Authorship Requirement

The Supreme Court established in Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), that copyright inheres in a human author's original intellectual conception. The Copyright Office formalized this position in its March 2023 guidance document, Copyright and Artificial Intelligence (U.S. Copyright Office, 2023), stating that the Office "will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author."

Registration as the Gatekeeper

Registration with the U.S. Copyright Office is not required for a copyright to exist, but it is required for a U.S. plaintiff to sue for infringement in federal court (17 U.S.C. § 411), and registration before infringement or within three months of first publication allows recovery of statutory damages up to $150,000 per willful infringement (17 U.S.C. § 504(c)(2)). When an applicant submits a work containing AI-generated material, the Copyright Office requires disclosure of that material and will exclude the AI-generated portions from the registration certificate.

The Zarya of the Dawn Precedent

In the February 2023 Zarya of the Dawn registration decision, the Copyright Office granted copyright only to the human-authored text and the selection and arrangement of images in a graphic novel — while explicitly excluding the individual AI-generated images produced by Midjourney. The images were excluded because the human author did not control "the specific expressive elements" of each image. This decision operationalized the Office's broader policy: human selection, arrangement, or modification of AI output can be protectable; the raw AI output itself is not.

Causal Relationships or Drivers

Three structural forces drive the current legal uncertainty.

1. Statutory Silence on Autonomous Systems

The Copyright Act of 1976 predates generative AI by decades. Congress drafted the statute around works produced by humans assisted by tools, not by systems that generate novel expression probabilistically. There is no provision — and no legislative history — addressing machine authorship. The Copyright Office has acted under existing statutory authority rather than waiting for congressional action.

2. The Thaler Litigation

Artist and AI researcher Stephen Thaler filed suit after the Copyright Office refused to register an image produced entirely by his Creativity Machine system. The U.S. District Court for the District of Columbia affirmed the denial in Thaler v. Perlmutter, No. 22-cv-01564 (D.D.C. Aug. 18, 2023), holding that human authorship is "a bedrock requirement of copyright" and that "non-human expression is not protected." Thaler's appeal to the D.C. Circuit is the most consequential pending litigation on this question as of the date of that ruling.

3. Training Data and Derivative Work Claims

Parallel litigation targets AI companies' use of copyrighted works to train models. Cases including Andersen v. Stability AI, No. 23-cv-00201 (N.D. Cal.), and Getty Images (US), Inc. v. Stability AI, No. 23-cv-00135 (D. Del.), raise questions about whether training constitutes reproduction under 17 U.S.C. § 106 and whether generated output constitutes an infringing derivative work. These cases sit at the intersection of AI output protection and AI system liability — a tension also relevant to AI liability under U.S. tort law.

Classification Boundaries

The Copyright Office's March 2023 guidance and subsequent registration practice establish four operative categories:

The critical variable in each category is whether the human claimant can demonstrate "creative control over expressive elements" — not merely that the human initiated or prompted the AI system. Prompting alone has not been found sufficient to constitute authorship, because the expressive choices in the output are made by the model, not the prompter.

Tradeoffs and Tensions

Incentive Structures vs. Public Domain Expansion

Copyright's constitutional basis — Article I, Section 8, Clause 8 — is utilitarian: to "promote the Progress of Science and useful Arts." Denying protection to AI-generated works means large volumes of expressive content enter the public domain immediately, which theoretically benefits the public but removes the incentive structure that traditionally encourages investment in creation. Whether AI companies require copyright protection to justify creative investment is contested.

Human Creativity Threshold Disputes

The line between "AI tool" and "autonomous AI author" is not self-defining. A photographer who selects a lens, aperture, and composition receives copyright despite the camera's mechanical execution. A user who selects a prompt and reviews output argues for an analogous position. The Copyright Office has rejected this analogy, but critics argue the threshold is applied inconsistently — a concern that intersects with questions about attorney ethics in AI use when lawyers rely on AI drafting tools.

International Divergence

The United Kingdom allows copyright for "computer-generated works" under the Copyright, Designs and Patents Act 1988, §9(3), vesting protection in the person who made the necessary arrangements for the work. China's Nanjing Internet Court reached a similar result in a 2023 ruling. U.S. refusal to protect AI output creates cross-border asymmetries relevant to licensing and enforcement — a dimension covered in the comparative treatment at international AI law vs. U.S. approaches.

Common Misconceptions

Misconception 1: Prompting an AI system makes the user the author.
The Copyright Office has explicitly stated that prompting does not constitute authorship if the user does not control the expressive elements of the output. The model's probabilistic generation process — not the user's prompt — determines the specific expression. Prompts function more like instructions to a ghostwriter than like creative acts in themselves.

Misconception 2: Registering AI-generated work without disclosure is a viable strategy.
Knowingly misrepresenting the nature of a work in a copyright application constitutes a violation that can void the registration (17 U.S.C. § 411(b)). The Copyright Office's 2023 guidance explicitly requires disclosure of AI-generated content. Omitting this disclosure exposes the registration to cancellation and potential fraud liability.

Misconception 3: Training a model on public domain works means the model's output is also public domain.
Public domain input status does not determine output status. Output is evaluated independently for human authorship. A model trained entirely on public domain text can still produce output that fails the human authorship test for different reasons.

Misconception 4: Copyright protection for AI output is settled law.
No appellate court has yet issued a definitive ruling. The Thaler D.C. Circuit appeal, the Andersen and Getty training data cases, and potential congressional action all remain unresolved. The current framework is administrative guidance and district-court-level precedent — persuasive but not binding nationally.

Checklist or Steps

The following sequence describes the analytical steps the Copyright Office and courts apply when evaluating a claim involving AI-generated material. This is a descriptive account of the regulatory and judicial process, not legal advice.

Step 1 — Identify all expressive elements in the work.
Distinguish portions that are textual, visual, musical, or structural. Note which elements were produced by a human independently of AI output.

Step 2 — Determine the source of each element.
For each element, establish whether a human made expressive choices (word selection, brush stroke, melodic phrase) or whether the AI system made those choices autonomously in response to input.

Step 3 — Evaluate prompt specificity and iterative control.
Document whether the human author revised, selected among, or modified AI outputs — and whether those acts of selection reflect original creative judgment rather than mere acceptance of machine output.

Step 4 — Assess whether AI-generated elements are separable.
If the work contains both human-authored and AI-generated elements, determine whether they can be identified and separated for purposes of a partial registration or disclaimer.

Step 5 — Prepare a disclosure statement for the Copyright Office.
Under the March 2023 guidance, applications must identify AI-generated content. The Office's online registration system includes a field for this disclosure.

Step 6 — File for registration of protectable elements only.
The Copyright Office will issue a certificate covering only the human-authored portions. The AI-generated portions are expressly excluded from the certificate's scope.

Step 7 — Monitor for administrative and legislative changes.
The Copyright Office is conducting a formal study on AI and copyright (Copyright Office Notice of Inquiry, 88 Fed. Reg. 59942, Aug. 30, 2023). The study's findings may prompt rulemaking or legislative referral. Existing registrations may be subject to re-examination if the policy framework shifts.

Reference Table or Matrix

AI-Generated Content: Copyright Framework Comparison

References

• U.S. Copyright Office — Copyright and Artificial Intelligence (2023)
• 17 U.S.C. § 102 — Subject Matter of Copyright
• 17 U.S.C. § 411 — Registration as Prerequisite to Infringement Suit
• 17 U.S.C. § 504 — Remedies for Infringement: Damages and Profits
• Copyright Office Notice of Inquiry: Artificial Intelligence and Copyright, 88 Fed. Reg. 59942 (Aug. 30, 2023)
• U.S. Copyright Office — Compendium of U.S. Copyright Office Practices, 3d ed.
• Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884) — Library of Congress
• U.S. District Court for D.C. — Thaler v. Perlmutter, No. 22-cv-01564 (Aug. 18, 2023)
• Federal Register — Copyright Office Rulemaking Notices
• U.S. Copyright Office — Title 17, Full Text