AI and Data Licensing · Memo

AI-Generated Work Product Ownership Under Work-for-Hire Doctrine After Recent USCO Guidance

The work-for-hire doctrine assumes a human author whose work product is contractually allocated to the hiring party. AI-assisted commissioned work strains that assumption. I am going to walk through how to allocate ownership when the underlying copyright status is contested.

The work-for-hire doctrine, codified at 17 U.S.C. section 101 and section 201(b), allocates copyright ownership of certain commissioned works to the hiring party rather than the author. The doctrine has two prongs: works prepared by an employee within the scope of employment, and specially commissioned works in nine enumerated categories where the parties have agreed in writing to the work-for-hire designation. The doctrine's operation has been stable for decades.

The United States Copyright Office's guidance in 2023-2025, including the Statement of Policy on AI-generated works and the three-part Report to Congress (Part 1 on Digital Replicas, Part 2 on Copyrightability, and Part 3 on Training), has clarified that purely AI-generated work product without sufficient human creative contribution is not copyrightable. The Thaler v. Perlmutter line of district court decisions has confirmed this position. The result is that some categories of commissioned work product may not generate copyrightable output at all, which destabilizes the work-for-hire allocation in any contract drafted assuming that copyright would attach.

What the USCO actually said

The USCO's Part 2 report addressed copyrightability of AI-assisted works. The position, in summary: works that consist entirely of AI-generated content, with no human creative selection, arrangement, or modification beyond the initial prompt, are not copyrightable. Works that include sufficient human creative contribution may be copyrightable, but only the human-authored elements are protected. Works that combine AI-generated content with human authorship may be registered with a disclaimer of the AI-generated portions.

The 'sufficient human creative contribution' standard is the source of most current uncertainty. The USCO has not provided a bright-line test. The case-by-case standard is whether the human's contribution involved sufficient creative expression to qualify as authorship under existing copyright doctrine. A prompt that merely directs the AI to produce an output is generally insufficient. Iterative human selection, arrangement, modification, and editing of AI outputs may be sufficient. The Thaler line confirms that purely autonomous AI output is not copyrightable; it does not draw the line for hybrid works.

Why this matters for the work-for-hire framework

The work-for-hire doctrine presupposes that the commissioned work is copyrightable. If the work is not copyrightable, there is no copyright to allocate; the work-for-hire designation has no operative effect on ownership of intellectual property rights.

This is the structural problem with most current AI-work-product contracts. The contract recites that the deliverables are works made for hire, that all right, title, and interest in the deliverables vest in the hiring party, and that the contractor assigns any residual rights to the hiring party. The recital assumes copyrightable output. If the output is not copyrightable, the recital allocates rights that do not exist. The hiring party receives the same access to the deliverables that anyone else would receive: no exclusivity, no infringement claim against subsequent users, no licensing leverage.

For commissioned work that is genuinely the product of human authorship using AI as a tool, the work-for-hire framework still operates as intended. The human author's contribution generates copyrightable expression; the work-for-hire designation allocates that copyright to the hiring party. The challenge is to identify, and contractually preserve, the human authorship contribution.

Drafting moves that preserve the work-for-hire framework

The contractual approach that I now use for AI-assisted commissioned work has several components.

First, the deliverables clause should require human creative contribution as a condition of acceptance. The contract specifies that the contractor's work must include identified categories of human contribution (selection, arrangement, modification, editing, judgment-based decisions) and that the contractor's invoices reflect the hours spent on those activities. The documentation supports the copyrightability claim if it is later questioned.

Second, the contract should require the contractor to document the development process. The contractor maintains a record of the prompts used, the AI outputs received, the human decisions made about those outputs, and the resulting deliverable. The record is the contemporaneous evidence of human authorship.

Third, the contract should address what happens if a particular deliverable turns out to be insufficiently copyrightable. The hiring party should have a contractual remedy (additional creative work to bring the deliverable above the copyrightability threshold, or a credit against future deliverables, or a refund of the fee for the non-copyrightable element).

The assignment-and-license fallback

Even with the documentation moves above, some deliverables will not generate copyright. The contract needs a fallback allocation for the non-copyright elements of the work. The standard structure:

1. Assignment of all rights. The contractor assigns to the hiring party any and all rights, including any copyright that may exist in the deliverables, any patent rights, any trade-secret rights, any moral rights, any neighboring rights, and any other intellectual property rights of any nature.
2. Joint ownership of non-copyright elements. If any elements of the deliverables are not protected by copyright but consist of valuable creative or technical contributions, the parties agree that the hiring party has exclusive use, modification, and distribution rights, with the contractor expressly waiving any rights to claim co-ownership or to assert competing claims.
3. Trade-secret treatment. The deliverables are designated as the hiring party's confidential and trade-secret information; the contractor agrees to maintain confidentiality and not to use the deliverables for any purpose outside the agreement. Trade-secret protection under the federal Defend Trade Secrets Act and California's Uniform Trade Secrets Act applies independently of copyright.
4. Non-compete or non-replication. The contractor agrees not to use the same prompts, methodologies, or creative direction to produce substantially similar deliverables for any other party for a defined period. The clause uses contractual restriction rather than copyright to protect the hiring party's investment.

This structure does not produce copyright where copyright does not exist. It does produce a defensible allocation of practical rights that the hiring party can use to protect its commercial interest in the deliverables.

Special considerations for the nine enumerated categories

The work-for-hire doctrine's specially commissioned category at 17 U.S.C. section 101 applies only to nine enumerated types of work: contributions to collective works, parts of motion pictures or audiovisual works, translations, supplementary works, compilations, instructional texts, tests, answer materials for tests, and atlases. Most modern AI-assisted commissioned work does not fall cleanly into any of these categories.

The categorical question matters because, for a specially commissioned work outside the nine categories, the work-for-hire designation has no statutory effect; copyright ownership remains with the author unless transferred by a separate written assignment. The contractual practice should not rely on the work-for-hire framework for work outside the nine categories. The assignment language is the operative provision.

The drafting tightening: every AI-assisted commissioned work contract should include both work-for-hire language and a separate present-tense assignment of all rights. The work-for-hire language operates where it can; the assignment operates where the work-for-hire framework does not.

The contractor-versus-employee distinction

The first prong of the work-for-hire doctrine, employee work within the scope of employment, has its own analytical framework after Community for Creative Non-Violence v. Reid, 490 U.S. 730 (1989). The Reid factors evaluate the parties' relationship to determine whether the worker is an employee under common-law agency principles. The factors include the hiring party's right to control the manner and means of work, the worker's skill, the source of instrumentalities, the location of work, the duration of the relationship, and the parties' tax treatment.

For traditional commissioned work, the Reid analysis is well-developed. For AI-assisted commissioned work, the analysis takes on new dimensions. The hiring party may provide access to the AI tools, may direct the prompts, may dictate the iteration process, and may make creative decisions about output selection. These elements may push the analysis toward employee status even where the worker is paid as a contractor on tax forms.

The drafting implication: the hiring party that wants to assert work-for-hire status under the employee prong should structure the relationship to support that characterization (consistent tax treatment, defined scope of work, ongoing relationship). The hiring party that prefers a contractor relationship for tax or labor-law reasons should rely on the specially-commissioned-work prong, where it applies, and on the assignment language otherwise.

International dimensions

The work-for-hire doctrine is a feature of United States copyright law. Many other jurisdictions, particularly civil-law systems, do not recognize work-for-hire allocation and instead require explicit assignment of rights or a license arrangement. Moral rights doctrines in countries including France, Germany, and others may persist with the original author regardless of contractual assignment.

For AI-assisted commissioned work with international elements (a contractor in one jurisdiction, a hiring party in another, deliverables intended for international use), the contract should address the international ownership question. The standard moves: a choice-of-law clause selecting United States law; an assignment of rights effective under each applicable jurisdiction's law; a waiver of moral rights to the extent permitted; and a representation by the contractor that the contractor has the authority to make the assignment under any applicable jurisdiction's law.

What I would not assume

The copyrightability of AI-assisted works is an actively evolving area of law. The USCO's guidance is current, but it operates against a developing body of district court and Court of Appeals decisions that may refine or reshape the analysis. Counsel drafting AI-assisted commissioned work contracts should treat the current state of the doctrine as a starting point and should build contractual protections that do not depend on any specific characterization holding up over time. The combination of work-for-hire language, present-tense assignment, trade-secret treatment, and contractual non-replication is more robust than reliance on any single allocation mechanism. The doctrine is genuinely murky at the line between human-authored AI-assisted work and AI-generated content with minimal human contribution; counsel should expect that the line will move as the case law develops. Outcomes in specific matters depend on the work, the development process, and the documentary record.

Sergei Tokmakov, Esq., CA Bar #279869. This memo is attorney commentary on legal questions and is not legal advice. Reading it does not create an attorney-client relationship. Past matter outcomes depend on facts and the responding party; nothing here is a prediction of result.