The question of who owns generative AI outputs in commercial settings sounds simple, but the legal answer is much more nuanced than many businesses expect. In ordinary conversation, people often assume that the person or company entering the prompt automatically owns whatever the system produces. In practice, however, ownership depends on several overlapping layers of law and contract that do not always align neatly. A business may control the output file, may receive contractual rights under a platform’s terms of service, may own the protectable human-authored aspects of a finished work, and may still face separate questions involving copyrightability, infringement, confidentiality, publicity rights, trade secrets, and regulatory compliance. For that reason, the most useful way to approach generative AI ownership is to separate possession from intellectual property, and commercial control from copyright ownership itself. ¹ ²
In the United States, the current legal baseline is that copyright protects original expression created by human authors. That principle matters because a work cannot be owned in the copyright sense unless it qualifies for copyright protection in the first place. The U.S. Copyright Office has taken the position that material generated entirely by AI, without sufficient human authorship, is not copyrightable, and the federal courts have reinforced that approach. The D.C. Circuit’s decision in Thaler v. Perlmutter underscored that the Copyright Act requires human authorship and upheld the refusal to register an image that the applicant had described as created autonomously by artificial intelligence. For commercial users, this means that a purely machine-generated output may still have business value, but it may not carry the exclusive legal protection that copyright normally provides. ¹ ³
That distinction is critically important because many companies confuse payment with ownership. Paying for access to a generative AI platform, paying an employee to craft prompts, or paying a consultant to supervise a creative workflow does not automatically create copyright in the final output. Copyright law first asks whether there is a legally recognized author. Only after that question is answered does the law address who owns the copyright. If the output is wholly machine-generated and lacks the required human authorship, there may be no copyright to own at all. A company may still possess the file, may still use it subject to platform terms, and may still incorporate it into a larger commercial project, but that is not the same thing as owning copyright in the raw generated material itself. ¹ ² ³
The position of the U.S. Copyright Office is especially significant because it does not treat every AI-assisted work as legally identical. The Office distinguishes between using AI as a tool in a human creative process and asking AI to create the expressive content itself. If a person uses an AI system in a way that assists human expression, copyright may still protect the human-authored aspects of the work. If, by contrast, the AI system determines the expressive details and the human contribution is too general, too remote, or too uncontrolled, the result becomes much harder to protect. This distinction matters because most commercial AI outputs are not purely autonomous artifacts. They are layered products involving prompts, source materials, editorial direction, selection, arrangement, revision, branding, and final human judgment. ¹
Prompting alone, however, is not a particularly strong ownership theory under current U.S. doctrine. The Copyright Office has explained that prompts generally function as instructions or requests rather than as authorship of the resulting expressive material. Even where prompting is detailed, iterative, and highly skilled, the final form of the output is often determined by the model’s own internal processes rather than by precise human control over expression. This is an uncomfortable point for many businesses because prompt engineering has quickly become a recognized commercial skill. Yet commercial value and legal authorship are not the same thing. A clever prompt may be valuable, and a prompt strategy may give a company a competitive advantage, but unless the final work reflects sufficient human-authored expression, prompting by itself will not reliably establish copyright ownership in the generated output. ¹
Where companies often have a stronger legal position is in the human contribution that surrounds, modifies, or structures the AI material. The U.S. Copyright Office has recognized that copyright may exist in human-authored content added to or retained within an AI-assisted work, and it may also exist in the original selection, coordination, or arrangement of otherwise unprotectable elements. This is often the most commercially important point. A raw AI-generated image, paragraph, slogan, or audio clip may not be protectable standing alone, but a finished advertising campaign, white paper, packaging design, investor deck, training manual, or illustrated report may contain enough human-authored structure and editorial judgment to qualify for protection. In those settings, the real intellectual property asset is often the final integrated deliverable rather than the model’s first draft. ¹
This leads to a practical commercial rule in the United States. If a company simply prompts a system to generate an image, article, or draft and then uses that result with little meaningful human revision, the company may have a useful asset but only a weak claim to exclusivity. If employees or retained creators instead shape the expressive content through original text, visual editing, sequencing, layout, adaptation, and substantial revision, then the final product is more likely to contain protectable human authorship. That difference affects far more than registration strategy. It affects licensing, valuation, due diligence, enforcement, indemnity, investor review, and the credibility of a company’s assertion that it owns its signature content. ¹ ² ³
Employment law then determines who owns the protectable human-authored portions once they exist. Under U.S. copyright law, copyright initially vests in the author, but works made for hire are treated differently. If an employee creates copyrightable material within the scope of employment, the employer is generally treated as the author and owner of the copyright unless the parties agree otherwise. In an AI-assisted commercial workflow, this means that a company will often own the human-authored edits, arrangements, or surrounding content created by its employees. But work-made-for-hire rules do not solve the threshold authorship problem. They allocate ownership of existing rights; they do not create copyright in material that never qualified for protection in the first place. If a work is entirely machine-generated and uncopyrightable, the employer cannot rely on work-for-hire doctrine to manufacture a copyright interest that the law does not otherwise recognize. ² ³
The analysis becomes even more contract-dependent when businesses use freelancers, agencies, consultants, or outsourced creative teams. Many commercial clients assume that paying an outside party for AI-assisted deliverables means they automatically own everything produced. That assumption is risky. If a contractor contributes copyrightable human expression, ownership may remain with the contractor unless the agreement clearly assigns those rights or qualifies under work-for-hire rules where legally applicable. Strong commercial drafting is therefore essential. Agreements should address ownership of human-authored deliverables, rights in prompts and workflows where appropriate, confidentiality of source materials, representations concerning third-party inputs, and documentation obligations that distinguish between machine-generated and human-created content. In many cases, careful contract language is what gives business practical control when copyright certainty is incomplete. ¹ ²
A comparative look at the United Kingdom shows that not every legal system approaches computer-generated works in exactly the same way. The Copyright, Designs and Patents Act 1988 states that in the case of a literary, dramatic, or musical, or artistic work that is computer-generated, the author is deemed to be the person by whom the arrangements necessary for the creation of the work are undertaken. The statute also defines a computer-generated work as one generated by computers in circumstances such that there is no human author. In addition, the Act provides that the author is generally the first owner of copyright, with an employer ordinarily becoming the first owner where the work is created by an employee in the course of employment. This statutory language gives the United Kingdom a more explicit framework for addressing certain computer-generated works than current U.S. doctrine provides. ⁴
Even so, the UK approach does not eliminate uncertainty. The phrase “the person by whom the arrangements necessary for the creation of the work are undertaken” may sound straightforward, but in modern generative AI systems it can be difficult to identify who actually made those arrangements. Was it the model developer, the platform provider, the enterprise user, the employee who designed the process, the creative director who refined the brief, or the individual who selected and finalized the output? In a collaborative business environment, several people may plausibly claim to have played that role. The statute offers a stronger textual basis for ownership arguments in some AI-related contexts, but it still leaves room for factual disputes and litigation about who should be treated as the relevant author. ⁴
The European Union adds another dimension to the discussion. The EU AI Act is not primarily a copyright ownership statute, but it has major implications for businesses commercializing generative AI outputs. The Regulation lays down obligations around transparency for certain AI-generated or manipulated content and imposes requirements on providers of general-purpose AI models, including copyright-related compliance measures and public summaries of training content. These rules do not answer the ownership question in the direct way that a copyright statute does, but they shape the commercial environment in which ownership claims are made. In the EU context, a company’s ability to use and market generative AI outputs may depend not only on authorship doctrine, but also on whether the underlying AI system and workflow satisfy emerging regulatory expectations about disclosure, traceability, and copyright-aware governance. ⁵
That regulatory environment has elevated the importance of provenance. In earlier technology markets, many buyers were content if a vendor simply delivered a usable product and warranted that it could be used. In the generative AI space, however, provenance is increasingly part of the value of the asset itself. Commercial counterparties want to know whether the final work was heavily revised by humans, whether sensitive internal materials were entered into the system, whether outputs are traceable to approved workflows, whether platform terms grant sufficient commercial rights, and whether the system provider maintains compliance measures relating to copyright and training data. These issues are not merely academic. They affect how confidently a company can license, sell, insure, or defend the outputs it places into the market. ¹ ⁵
It is also crucial to understand that ownership and non-infringement are not the same question. A business may own protectable human-authored portions of an AI-assisted work and still face legal exposure if the final output reproduces protected expression from another source, misuses confidential or proprietary inputs, imitates a person’s voice or likeness, or creates confusion with a third party’s brand. Ownership tells a company what rights it may have in its own contribution. It does not automatically establish that the work is safe to use or that no one else has a competing claim. For commercial actors, this means that internal AI governance should never stop at the question, “Do we own it?” The better question is, “What do we own, what are we allowed to use, what risks remain, and what evidence supports our position?” ¹ ³ ⁵
Another practical issue in commercial environments is the difference between internal and external use. A company may be comfortable using a lightly edited AI output internally for brainstorming, internal presentations, or temporary concept development because the legal and commercial stakes are relatively lower. The analysis changes when the same material becomes customer-facing advertising, licensed educational content, software documentation, packaging, public thought leadership, or a branded campaign distributed at scale. The more visible and valuable the output becomes, the more important it is to establish who contributed what, what rights were assigned, what systems were used, and whether the final asset contains a defensible layer of human authorship. In other words, the market significance of the work often determines how much ownership precision a company needs. ¹ ² ⁵
This is why businesses should think of generative AI ownership as a documentation problem as much as a legal problem. A company is in a far stronger position if it can show drafts, editorial changes, revision history, human-created source text, internal approval records, and contracts that allocate rights clearly. That kind of record can help support a later copyright application, strengthen a licensing negotiation, reassure investors or acquirers during diligence, and reduce the chances of confusion about whether the final work was mostly machine-generated or substantially shaped by humans. In commercial disputes, the side that can explain the workflow with precision often has a practical advantage over the side that can only say that an employee used AI and produced a file. ¹ ² ³
The same logic applies to platform terms. Many businesses assume that if an AI provider says users’ “own” outputs, the legal issue is settled. In reality, platform language typically allocates contractual rights between the user and the provider, but it does not override copyright statutes or guarantee that an output is copyrightable against the rest of the world. A platform can agree not to assert its own claims, can assign whatever rights it may have, and can grant broad usage permissions, but it cannot create statutory copyright in uncopyrightable material simply by contract wording. Commercial users should therefore read output ownership clauses carefully and understand that such provisions are important, but they answer only one part of a larger legal question. ¹ ²
For that reason, the best commercial approach is usually layered. A business should aim to secure contractual rights from the platform, obtain assignments from contractors where needed, ensure that employees create and revise work within documented workflows, preserve evidence of human authorship, and review outputs for infringement and compliance risks before large-scale distribution. None of these steps alone guarantees exclusive ownership of every AI-assisted asset, but together they create a much stronger legal and commercial position. Generative AI does not eliminate the traditional disciplines of intellectual property management. Instead, it makes those disciplines more important because the boundaries of authorship are more contested and more dependent on facts. ¹ ² ³ ⁵
So who owns generative AI outputs in commercial contexts? The most defensible answer is that ownership depends on what exactly is being claimed. In the United States, a purely autonomous AI output may not attract copyright protection at all because current doctrine requires human authorship. A business may nevertheless own the protectable human-authored elements added through selection, revision, arrangement, commentary, branding, or integration into a larger final work. It may also control the output through employment law, assignment agreements, license terms, secrecy, and platform contracts. In the United Kingdom, statutory language may allow stronger arguments for protection of certain computer-generated works, although factual uncertainty remains. In the European Union, the ownership question increasingly sits beside a growing compliance framework that businesses must take seriously when commercializing generative AI. ¹ ² ³ ⁴ ⁵
The commercial lesson is therefore narrower, but far more useful, than the popular debate often suggests. Generative AI does not create a new world in which ownership automatically belongs to the prompt writer, nor does it eliminate the need for conventional intellectual property analysis. Instead, it requires businesses to be more disciplined about authorship, documentation, contracts, workflow design, and regulatory awareness. The companies that will be best positioned in the market are not merely those that generate the most content. They are the ones that can explain, with evidence, which parts of a final delivery are human-authored, which rights were secured by contract, which systems were used, and why the business can confidently commercialize the result. In that sense, ownership of AI creations is not a single yes-or-no issue. It is an issue of structure, proof, and governance. ¹ ² ³ ⁵
This article is for general informational purposes only and does not constitute legal advice. ¹ ³
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).
References
- U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (January 29, 2025). www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
- 17 U.S.C. § 201, Ownership of Copyright. https://www.law.cornell.edu/uscode/text/17/201
- Thaler v. Perlmutter, United States Court of Appeals for the District of Columbia Circuit, No. 23-5233, decided March 18, 2025. https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf
- Copyright, Designs and Patents Act 1988 (United Kingdom), especially sections 9, 11, and 178. https://www.legislation.gov.uk/ukpga/1988/48/contents
- Regulation (EU) 2024/1689, Artificial Intelligence Act, including provisions on transparency and general-purpose AI model obligations. https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng