Who Owns the Copyright in Work Generated by an LLM?

Key Takeaways

• LLM-generated works raise copyright ownership questions because U.S. copyright protection still depends on human authorship, and current guidance treats autonomously generated AI content differently from human-created or human-edited work.
• Prompting an LLM may not establish authorship of the resulting expression, which can leave valuable company content without federal copyright protection even if a platform contract says the customer owns the output.
• Companies using LLMs for important publications or proprietary materials should preserve meaningful human creative involvement and documentation showing who made the expressive choices reflected in the final work.

Large language models (LLMs) are now routine business tools. Lawyers use them to draft client communications. Companies use them to create advertising copy or prepare internal reports. Their growing use raises a question that sounds simple but is not: Who owns the copyright in something generated by an LLM?

Under current U.S. law, the starting point is clear. Copyright requires human authorship. An artificial intelligence system cannot own a copyright because it cannot be an author under the Copyright Act. The more difficult question is whether the human using the LLM contributed enough creative expression to become the author of some or all of the resulting work.

Copyright Begins With a Human Author

Section 201 of the Copyright Act provides that copyright initially belongs to the author of a protected work. In a qualifying work made for hire, the employer is treated as the author and generally owns the copyright. But those ownership rules matter only if copyright exists in the first place.

The dividing line became clearer in Thaler v. Perlmutter. Stephen Thaler sought copyright registration for an image he represented as having been generated autonomously by an AI system. In March 2025, the D.C. Circuit held that the Copyright Act requires eligible works to be authored initially by a human being. 130 F.4th 1039.

Thaler asked the Supreme Court to review that decision. On March 2, 2026, the Court denied certiorari. The denial does not constitute Supreme Court approval of the D.C. Circuit’s reasoning, but it leaves the decision intact.

Importantly, Thaler involved a work represented as autonomously created by AI. It did not decide the harder question confronting most companies: How much human involvement is enough when AI is used as part of the creative process?

A Prompt Is Not Necessarily Authorship

The Copyright Office (the “Office”) addressed that question in Part 2 of its Copyright and Artificial Intelligence report. Its position is that AI-assisted works can receive copyright protection when a human determines sufficient expressive elements of the resulting work. Human modifications to AI-generated material may qualify. So may creative human selection or arrangement. Prompts, however, present a harder issue.

The Office concludes that, with current generally available technology, prompts alone ordinarily do not give users enough control over the expressive elements of an AI output to make them authors of that output. A person may provide detailed instructions and receive precisely the type of document requested. That does not necessarily mean the person authored the language the LLM generated. As the Office noted, “prompts may reflect a user’s mental conception or idea, but they do not control the way that idea is expressed.”

Consider an employee who tells an LLM: “Write an article explaining our company’s new product to customers,” then publishes the response with only minor corrections. The company should not assume that it owns a copyright in the AI-generated language. Under the Office’s present approach, some or perhaps most of that language may have no federal copyright protection. That has a significant business consequence. Competitors may be able to copy unprotected AI-generated expression without infringing the company’s copyright.

Human Editing Can Change the Result

The analysis changes when the LLM assists a human creator rather than replacing one. Suppose an executive writes an article and uses an LLM to suggest edits. She chooses which suggestions to accept and rewrites portions herself. The human-created expression remains eligible for protection. Her new creative revisions may also be protected.

The question is therefore not simply whether AI was used. The question is who made the expressive choices reflected in the final work.

The Congressional Research Service (CRS) describes much the same framework. AI-assisted works may contain copyrightable human contributions even though the portions generated by the machine are excluded from protection. The Office reports that it has registered hundreds of works incorporating AI-generated material while limiting protection to the human-authored portions.

A pending case could provide more guidance. In Allen v. Perlmutter, artist Jason Allen is challenging the Office’s refusal to register an image created through extensive interaction with Midjourney followed by human editing. No. 24-cv-02665-WJM (D .Colo. 2024). As of August 2026, the litigation remains pending. The case matters because it may help define when human direction of an AI system becomes human authorship.

Who Owns AI-Assisted Work Inside a Company?

For in-house lawyers, authorship and ownership are separate questions.

If an employee produces copyrightable human expression within the scope of employment, the work made for hire doctrine will often place ownership with the employer. An outside contractor presents a different issue because commissioned works qualify as works made for hire only under specified circumstances. Written copyright assignments therefore remain important. The LLM itself owns nothing under current U.S. copyright law.

The terms of the AI platform also deserve attention. A provider may contractually assign whatever rights it possesses in an output. But a contract cannot manufacture a federal copyright where copyright law recognizes no human authorship. A provision saying that the customer “owns the output” does not necessarily mean the output is protected against copying by others. For valuable corporate content, that distinction can be critical.

Ownership Does Not Eliminate Infringement Risk

There is another issue companies should keep separate from ownership. An AI-generated work may lack copyright protection for the user while still containing expression protected by someone else’s copyright.

The CRS report recognizes that AI outputs remain subject to traditional infringement principles. If an output copies protectable expression from an existing work and the legal requirements for infringement are satisfied, use of the output may create liability even though the company itself has no copyright in it. Thus, saying “the company owns the AI output” does not resolve the copyright analysis.

What Companies Should Do

Companies do not need to avoid LLMs. They should, however, build copyright considerations into how valuable content is created.

Where copyright protection matters, meaningful human creativity should remain part of the process. Companies should also preserve enough information to show the human contribution to the final work. For important publications or proprietary materials, relying on a prompt followed by minimal editing creates greater uncertainty about whether copyright protection exists.

The law is still developing. Allen may help clarify the line between human creative control and machine generation. Congress could eventually address the issue as well.

For now, the most useful rule is simple:

Using an LLM does not prevent copyright protection, but receiving an AI-generated answer does not create copyright ownership. Copyright follows human authorship. Once sufficient human authorship exists, ordinary rules governing employment and contractual transfers determine who owns it.

For questions about the issues discussed in this alert, contact your Polsinelli attorney.