Artificial intelligence is changing how independent contractors create software, marketing materials, research reports, designs, business strategies, audiovisual content, and other valuable work product. A contractor who once drafted an article, designed a logo, or wrote source code manually may now use generative AI to produce an initial draft, develop alternatives, correct errors, translate material, summarize data, generate images, or complete portions of a larger deliverable. These tools can improve efficiency, but they also complicate a question that businesses have often treated as routine: Who owns the finished work?
The answer cannot safely be determined from the fact that the business paid for the project. Payment alone does not necessarily transfer copyright ownership. Nor does labeling a person an “independent contractor” resolve the issue. In many circumstances, the contractor initially owns the copyright in the contractor’s human-authored contributions unless the parties execute an enforceable written assignment. The situation becomes even more complicated when portions of the deliverable were generated by AI and may not qualify for copyright protection at all.
A business engaging an independent contractor must therefore address several distinct subjects. It must determine whether the contractor or the business owns the human-authored portions of the deliverable, whether any AI-generated portions are legally protectable, whether the contractor had authority to submit business information to the AI system, whether the output incorporates third-party content, and whether the contractor has transferred all associated intellectual-property rights. These questions should be answered before work begins rather than after a product launch, financing transaction, acquisition, or ownership dispute.
Many businesses assume that commissioning and paying for a deliverable automatically gives the business all rights in it. That assumption may be commercially understandable, but it is frequently incorrect under federal copyright law.
Copyright ordinarily belongs initially to the author of an original work. When an employee creates a work within the scope of employment, the Copyright Act generally treats the employer as the author under the work-made-for-hire doctrine. Independent contractors are treated differently. A commissioned work created by an independent contractor qualifies as a work made for hire only if it falls within one of the statutory categories and the parties expressly agree in a signed writing that the work will be considered a work made for hire. ¹
Those statutory categories are limited. They include works commissioned for use as contributions to collective works, parts of motion pictures or other audiovisual works, translations, supplementary works, compilations, instructional texts, tests, answer material for tests, and atlases. Many common contractor deliverables do not comfortably fit within those categories. A custom software application, standalone website design, business plan, logo, internal procedure, sales presentation, or independently commissioned article may not become a work made for hire merely because the contract uses that phrase. ³
The United States Supreme Court emphasized this distinction in Community for Creative Non-Violence v. Reid. The Court held that courts should use common-law agency principles to determine whether a creator is an employee for copyright purposes. Relevant circumstances include the hiring party’s right to control the work, the skill required, the source of tools, the location of the work, the duration of the relationship, the method of payment, the provision of benefits, the tax treatment of the creator, and other aspects of the relationship. No single contractual label necessarily controls. ²
This means that a business can face an ownership problem from either direction. A person labeled as an independent contractor may actually qualify as an employee under the applicable legal test. Conversely, a contractor whom the business closely supervises may still remain an independent contractor whose work does not qualify as a statutory work made for hire. The agreement should not depend exclusively on a classification label or on work-made-for-hire language. It should contain a present, express assignment of all intellectual-property rights as a contractual backstop.
Before AI became common in professional work, a contractor deliverable could usually be analyzed by identifying the human author and determining whether that person had transferred the resulting rights. AI-assisted deliverables require a more layered analysis.
A single deliverable may contain the client’s preexisting information, the contractor’s original writing or design, material generated by an AI system, third-party content identified by the AI system, modifications made by the contractor, and a creative arrangement of human-authored and AI-generated elements. Each layer can have a different legal status.
For example, a marketing contractor may receive a company’s product specifications, customer information, and brand guidelines. The contractor may use those materials to create prompts, generate multiple advertising concepts, rewrite selected passages, add original slogans, combine AI-generated images with human-created graphics, and arrange the final material into a campaign. The business may own its original specifications and brand assets. The contractor may initially own the contractor’s original expression. The AI-generated portions may be unprotectable. Certain third-party elements may be subject to another owner’s rights. The overall selection and arrangement may be protectable if it reflects sufficient human creativity.
A contract that simply states that “the client owns the final deliverable” may not adequately address these layers. The agreement should define work product broadly enough to include drafts, prompts, prompt sequences, system instructions, source code, scripts, workflows, configurations, fine-tuning materials, datasets, research notes, output variations, edits, documentation, and derivative materials. It should also distinguish between newly created work product, the contractor’s preexisting tools, the client’s preexisting materials, third-party materials, and AI-generated content.
This distinction matters because a contractor cannot assign rights that the contractor does not possess. If a portion of an output is not copyrightable, the contractor cannot create copyright protection merely by assigning it. If a third party owns incorporated material, the contractor cannot transfer complete ownership without the necessary authorization. If the contractor uses preexisting code or templates, the business may receive only a license unless the agreement provides otherwise.
The central copyright principle governing AI-generated material is that copyright protects human authorship. In 2025, the United States Court of Appeals for the District of Columbia Circuit held in Thaler v. Perlmutter that the Copyright Act requires a work to be authored in the first instance by a human being. The case involved an image that the applicant represented had been created autonomously by an AI system, with the system identified as the sole author. The court affirmed the denial of registration. In March 2026, the United States Supreme Court denied the petition seeking review, leaving the appellate decision in place.⁵
That ruling does not mean that every work made with AI is ineligible for copyright protection. The D.C. Circuit distinguished between a work created entirely by a machine and a work created by a human using AI as an assisting tool. The difficult question is not whether AI appeared somewhere in the creative process. It is whether the protectable expression in the final work resulted from sufficiently creative human choices.
The United States Copyright Office has taken the position that prompts alone, using currently available general-purpose technologies, ordinarily do not give the user enough control over the expressive elements of the resulting output to make the user the author of that output. A detailed prompt may describe the desired subject, style, mood, or composition, but the system may independently determine the particular expression that appears in the result. The Office therefore evaluates the human contribution rather than assuming that the person who entered the prompt authored everything the system produced.⁴
Human modifications can nevertheless qualify for protection. A person may rewrite AI-generated text, alter an AI-generated image, create original code around generated code, or make other changes that independently satisfy the minimal originality requirement. Copyright may then protect the human-authored additions even though it does not protect the underlying machine-generated material.
The same principle applies to selection and arrangement. A person may creatively select, coordinate, and arrange AI-generated and human-created material into a protectable compilation. In that situation, copyright may cover the original organization of the overall work without extending to the unprotectable AI-generated elements standing alone.⁴ The Copyright Office’s registration guidance similarly directs applicants to identify human-authored contributions and exclude more-than-minimal AI-generated material from the claim.⁶
For businesses, this creates a practical ownership limitation. Even a perfectly drafted assignment may transfer only the copyright that exists. If a contractor delivers a largely autonomous AI output with little protectable human authorship, the business may own the contractual rights to use the output but possess little or no copyright capable of preventing competitors from copying it. The company may therefore receive exclusivity by contract against the contractor while lacking exclusivity against the rest of the market.
Many independent-contractor agreements include a sentence stating that all deliverables are “works made for hire.” That language is useful, but it should not be the only mechanism used to transfer ownership.
A commissioned contractor work qualifies as a work made for hire only when the statutory requirements are met. The work must fall within one of the recognized categories, and the parties must expressly agree in a signed writing that it will be treated as a work made for hire. ¹ The Copyright Office describes four conditions for a specially ordered or commissioned work, including eligibility under one of the nine statutory categories and a signed agreement. ³
Consider a business that hires a contractor to develop an AI-enabled customer portal. The contract may call the portal a work made for hire, but standalone computer software is not expressly listed among the nine commissioned-work categories. Depending on the facts, the software might be characterized as part of a collective work or compilation, but relying on that characterization creates unnecessary uncertainty. A direct assignment of copyright rights is more dependable.
The better formulation generally states that the work will be treated as a work made for hire to the maximum extent legally permitted and that, to the extent any portion does not qualify, the contractor presently assigns all right, title, and interest in that work to the client. The assignment should be written as a present transfer rather than merely a promise to transfer rights later. Language stating that the contractor “hereby assigns” is ordinarily preferable to language stating only that the contractor “will assign.”
The agreement should also authorize the business to register, maintain, enforce, and defend the transferred rights. It should require the contractor to execute confirmatory assignments and other documents after the project ends. A power-of-attorney provision may be appropriate when narrowly drafted for situations in which the contractor fails to cooperate, although its enforceability and scope should be reviewed under the governing law.
Section 204 of the Copyright Act generally requires a transfer of copyright ownership to be memorialized in a writing signed by the owner of the rights conveyed. ¹ Informal understandings, invoices, payment records, or oral assurances may not provide the clarity required for a complete transfer. A business should obtain the signed assignment before the contractor begins producing valuable work or, at the latest, before final payment.
AI-related agreements should be drafted around human contributions rather than treating an AI system as an employee, contractor, or legal author. The D.C. Circuit rejected the argument that the work-made-for-hire doctrine could transform a machine into an author. The doctrine determines who owns a copyright in qualifying human-authored work; it does not eliminate the requirement that the work originate with a human author.⁵
An effective agreement should therefore assign the contractor’s human-authored contributions to the work product, including revisions, selection, arrangement, editing, annotations, original prompts, code modifications, and other protectable expression. It should separately allocate contractual control over AI-generated material, even when that material may not qualify for copyright protection.
Contractual control can still have significant value. The contractor can agree not to reuse, sell, publish, distribute, or provide substantially similar project-specific output to another customer. The contractor can agree to deliver all generated alternatives and delete retained copies. The contractor can grant the client all transferable rights provided by the AI vendor’s terms. The contractor can also promise not to challenge the client’s ownership or authorized use.
These contractual restrictions do not create copyright where none exists, and they may not bind third parties who independently obtain or recreate the same material. They can, however, prevent the contractor from exploiting the deliverable in a manner inconsistent with the business relationship. They can also support breach-of-contract and confidentiality claims if the contractor misuses project materials.
In many AI-assisted projects, the most valuable work product is not the visible output. It is the process used to produce reliable outputs. That process may include proprietary system prompts, prompt chains, model configurations, retrieval instructions, source-selection criteria, data-cleaning rules, human-review procedures, evaluation rubrics, testing protocols, and methods for correcting recurring errors.
A contractor agreement should expressly address these materials. Without clear language, a contractor may argue that prompt libraries, reusable code, templates, and workflows are part of the contractor’s general knowledge or preexisting toolkit rather than client-owned work product. The client may believe it purchased a complete operational system while the contractor believes it delivered only a limited result.
The appropriate allocation depends on the transaction. A contractor who uses a longstanding proprietary framework for many customers may reasonably retain ownership of that framework while granting the client a sufficiently broad license. By contrast, a workflow designed specifically from the client’s confidential information, funded by the client, and intended exclusively for the client’s business may appropriately belong to the client.
The agreement should require the contractor to identify preexisting materials before incorporating them into the project. It should state that undisclosed preexisting materials are either assigned with the work product or licensed to the client on broad, irrevocable terms. The license should be sufficient to permit the client and its successors, service providers, purchasers, and affiliates to use, reproduce, modify, maintain, and commercialize the finished product without returning to the contractor for further permission.
Ownership is only one part of the risk. Contractors frequently receive information that the business treats as confidential, including source code, customer data, pricing information, product roadmaps, financial projections, legal documents, technical specifications, and internal procedures. Entering that information into an external AI system may create disclosure, retention, access, or reuse risks.
Under the federal Defend Trade Secrets Act, information generally qualifies as a trade secret only when the owner has taken reasonable measures to keep it secret and the information derives independent economic value from not being generally known or readily ascertainable through proper means.⁷ Michigan’s Uniform Trade Secrets Act contains similar principles and provides remedies for the misappropriation of qualifying trade secrets.⁸
A company that permits unrestricted contractor use of public AI tools may later encounter arguments about whether it took reasonable measures to protect its information. The result will depend on the facts, including the sensitivity of the information, the tool’s settings, the vendor’s contract, retention practices, access controls, and the company’s policies. Nevertheless, businesses should not assume that a conventional confidentiality clause adequately addresses AI use.
The agreement should state whether the contractor may use generative AI at all. If use is allowed, it should identify approved systems, accounts, security settings, data classifications, and permitted purposes. The contractor should be prohibited from entering confidential information, personal information, privileged communications, trade secrets, export-controlled information, or regulated data into an unapproved system.
The contractor should also be required to comply with the client’s information-security policies, notify the client of unauthorized disclosure, preserve relevant records, cooperate with incident-response efforts, and delete client information at the end of the engagement. Where appropriate, the business should require the contractor to use an enterprise account configured so that inputs and outputs are not used to train generally available models.
Confidentiality clauses should cover not only materials supplied by the client but also project-specific prompts, intermediate outputs, evaluations, errors, testing data, and observations about the client’s systems. A contractor’s prompt may reveal confidential business information even when the prompt does not reproduce an entire source document.
An assignment from the contractor does not eliminate the possibility that the deliverable infringes another party’s rights. AI-generated output may resemble existing text, software, images, music, characters, or other protected material. A contractor may also intentionally place third-party content into a prompt, use reference materials without authorization, or incorporate open-source code without observing the applicable license.
The agreement should require disclosure of third-party and open-source materials used in the project. It should prohibit the contractor from incorporating material that would require the client to disclose proprietary source code, license the finished product on unwanted terms, pay undisclosed royalties, or impose attribution requirements inconsistent with the intended use.
Representations and warranties should be calibrated to the nature of the project. A contractor may reasonably represent that the contractor has not knowingly copied protected material and has complied with specified procedures. A sophisticated development firm may be expected to provide stronger warranties regarding originality, licensing, and noninfringement. A contractor using probabilistic generative systems may resist an absolute guarantee that no portion of any output resembles any existing work.
The parties can address this tension through process-based obligations. The contractor may be required to use approved tools, avoid prompts requesting imitation of living artists or identifiable third-party works, conduct code and content reviews, use similarity-detection procedures, maintain records of source materials, and obtain approval before incorporating external content. These requirements do not eliminate risk, but they create a defensible process and help allocate responsibility when a claim arises.
A contractor agreement cannot transfer more rights than the contractor obtained under the applicable AI vendor’s terms. Some providers purport to assign or disclaim rights in outputs, while others impose restrictions relating to account type, content, model improvement, prohibited uses, indemnification, or publication. These terms can change, and different terms may apply to free accounts, consumer accounts, application programming interfaces, and enterprise services.
The contractor should be required to identify the AI systems used and confirm that the intended commercial use complies with the governing terms. The agreement should prevent a contractor from using a personal or free account when doing so would expose client information, restrict commercial use, or provide weaker contractual protections than an approved enterprise account.
The client should also consider whether the contractor’s use of the tool is consistent with the client’s own obligations. A company may have contractual duties to its customers concerning confidentiality, data location, cybersecurity, subcontractors, or automated processing. Permitting a contractor to transmit customer data to an undisclosed AI provider could violate those duties even if the output itself is commercially useful.
Because vendor terms and technical settings evolve, the agreement should not depend solely on a representation made at the beginning of a long project. The contractor should have a continuing duty to notify the client of material changes affecting ownership, confidentiality, data use, output restrictions, or indemnification.
AI-assisted work can also produce potentially patentable inventions. Patent law, like copyright law, currently requires human inventorship. In Thaler v. Vidal, the United States Court of Appeals for the Federal Circuit held that an inventor under the Patent Act must be a natural person.⁹
The use of AI does not necessarily disqualify an invention from patent protection. The key issue is whether one or more humans made a legally sufficient contribution to the conception of the claimed invention. That question can be fact-intensive, especially when a contractor uses AI to generate alternatives, identify technical solutions, or optimize a design.
A contractor agreement should therefore include an invention-assignment provision in addition to a copyright assignment. It should require disclosure of potentially patentable developments, cooperation with patent applications, execution of assignments, and preservation of records showing the human contribution to conception. The business should not assume that a copyright clause automatically transfers patent rights, trade-secret rights, database rights, or other forms of intellectual property.
Records can be especially important. Prompt histories, design notes, test results, human revisions, rejected alternatives, and communications may help establish who contributed to the inventive concept. At the same time, retaining unnecessary prompt histories can create confidentiality and litigation risks. The parties should adopt a deliberate record-retention protocol rather than allowing each contractor to decide independently what to preserve.
Ownership clauses determine who receives rights, but they do not fully allocate the cost of a third-party claim. A comprehensive agreement should address representations, indemnification, defense obligations, limitations of liability, and insurance.
The contractor may represent that the contractor has authority to enter the agreement, owns or controls all assigned human-authored contributions, has disclosed preexisting and third-party materials, has complied with approved AI-use procedures, and has not knowingly submitted restricted client information to an unauthorized system. The contractor may also represent that no prior agreement prevents the transfer of rights.
Indemnification language should identify the claims covered, who controls the defense, how settlements are approved, and whether the obligation applies to intellectual-property infringement, confidentiality breaches, privacy violations, data-security incidents, and violations of AI vendor terms. An indemnity that covers infringement but excludes all consequential losses or is subject to a nominal liability cap may provide limited practical protection.
The parties should consider whether a separate or elevated liability cap applies to intellectual-property and confidentiality claims. The client should also determine whether the contractor carries technology errors-and-omissions, cyber, or media-liability coverage appropriate to the work. Insurance does not replace careful drafting, and policy exclusions must be reviewed, but it can provide a meaningful source of recovery.
A business that receives AI-assisted work should be able to explain how the work was created. That does not require preserving every keystroke or disclosing proprietary methods publicly. It does require sufficient documentation to distinguish human-authored expression from machine-generated material and to substantiate the chain of title.
The contractor should document the tools used, the nature of AI involvement, the human modifications made, the sources of third-party materials, and the assignment of rights. For valuable works, the company may also require delivery of editable source files, prompt histories, code repositories, design files, model configurations, and revision records.
Accurate documentation becomes especially important when seeking copyright registration. Applicants must disclose more-than-minimal AI-generated content and identify the human-authored material being claimed.⁶ Failure to make an accurate disclosure can delay registration, narrow the claim, or complicate later enforcement.
Documentation is also important during investments, lending transactions, and acquisitions. A buyer evaluating a software company or creative business may ask whether contractors signed assignments, whether AI was used, whether training data or customer information was submitted to external systems, and whether important deliverables contain unlicensed third-party materials. Missing assignments can reduce valuation, delay closing, or require former contractors to be located years after the work was completed.
Michigan businesses do not need an entirely separate contract for every project involving AI, but their independent-contractor agreements should be updated to account for AI-assisted work. The provisions should be tailored to the nature of the engagement. A contractor writing internal meeting summaries presents different risks from a developer creating a customer-facing software platform or a designer producing a company’s principal brand assets.
The agreement should define the deliverables and work product with precision, include work-made-for-hire language where potentially applicable, and contain a present assignment of all transferable rights. It should allocate ownership of prompts, workflows, code, data, and improvements. It should identify retained contractor materials and provide a sufficient license for anything embedded in the deliverable.
The agreement should also regulate AI use rather than leaving the subject unaddressed. Depending on the project, the client may prohibit AI, require advance written approval, permit only approved tools, or allow broader use subject to confidentiality and review requirements. The contractor should disclose material AI use and remain responsible for the quality, accuracy, legality, and security of the final work.
Finally, the agreement should coordinate intellectual-property provisions with confidentiality, data security, privacy, indemnification, termination, record retention, and dispute-resolution clauses. Ownership provisions cannot be isolated from the rest of the contract. A client that owns a deliverable but lacks the source files, necessary licenses, passwords, documentation, or legal right to use embedded components may not have obtained the practical benefit it expected.
AI has not eliminated the need for traditional intellectual-property agreements. It has made careful contracting more important. The most significant risk is not that an AI system will unexpectedly claim ownership. Current United States law does not recognize an AI system as a copyright author or patent inventor. The greater risk is that the parties will misunderstand what was created, who contributed protectable human authorship, what rights the contractor possessed, and what information was disclosed during the process.
Businesses should not rely solely on payment, work-made-for-hire terminology, or a general statement that the client owns the deliverables. Independent-contractor agreements should include an express assignment of human-authored contributions, contractual control over AI-generated material, restrictions on the use of confidential information, disclosure of AI tools and third-party content, and continuing cooperation obligations.
The goal is not merely to state that the business owns the final output. The agreement should establish a defensible chain of title, preserve confidentiality, permit the intended commercial use, and reduce the likelihood that the company will later discover that a valuable asset is unprotectable, unlicensed, or still owned by the contractor who created it.
This article is intended for general informational purposes and does not constitute legal advice. The appropriate contractual provisions depend on the nature of the project, the technology used, the information involved, and the governing law.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
Sources
1- 17 U.S.C. §§ 101, 201, and 204. https://www.law.cornell.edu/uscode/text/17/101
2- Community for Creative Non-Violence v. Reid, 490 U.S. 730 (1989). https://supreme.justia.com/cases/federal/us/490/730/
3- United States Copyright Office, Circular 30: Works Made for Hire. https://www.copyright.gov/circs/circ30.pdf
4- United States Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, January 2025. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
5- Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025), cert. denied, No. 25-449, March 2, 2026. https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf
6- United States Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190, March 16, 2023. https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence
7- Defend Trade Secrets Act, 18 U.S.C. §§ 1836 and 1839. https://www.law.cornell.edu/uscode/text/18/1836
8- Michigan Uniform Trade Secrets Act, MCL 445.1901–445.1910. https://www.legislature.mi.gov/Laws/MCL?objectName=mcl-445-1901
9- Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). https://www.sternekessler.com/news-insights/insights/thaler-v-vidal-43-f4th-1207-fed-cir-2022-moore-taranto-stark/
10- Feist Publications, Inc. v. Rural Telephone Service Co., 499 U.S. 340 (1991). https://supreme.justia.com/cases/federal/us/499/340/