Most creators using AI tools right now believe they own the work the tool produces. The law says they do not. Not for the portions the AI authored. The line between what you own and what you don't is thinner than most realize, and where you fall on it determines whether you have any legal remedy if someone takes your work.
This is not a policy debate about what copyright law should do. It is a description of where copyright law stands today.
Copyright has always required a human author
The human authorship requirement is not new. It is not a response to AI. It is the foundational principle of American copyright law, grounded in the Copyright Clause of the Constitution and present in every major copyright statute since 1790.
Copyright protects original works of authorship. Authorship, under U.S. law, has always meant human authorship. The Copyright Office has required this for over 150 years. Courts have enforced it consistently.
What is new is the number of creators producing work with AI tools, the volume of that work being published and monetized, and the urgency of understanding exactly what that work is worth legally.
The cases that reached the courts did not change the rule. They applied it.
Stephen Thaler built an AI system called DABUS and attempted to register the work it generated, listing the AI as the author. The Copyright Office refused the registration. The federal district court affirmed the refusal. The U.S. Court of Appeals for the D.C. Circuit affirmed it again. The Supreme Court declined to hear the case. The result is settled: an AI cannot be an author under U.S. copyright law.
What the Copyright Office determined about AI-generated output
In January 2025, the U.S. Copyright Office published Part 2 of its Report on Copyright and Artificial Intelligence. The central conclusion: works solely generated by AI are not copyrightable.
This is a statutory interpretation, not a policy preference. The Copyright Act protects "works of authorship." Authorship requires a human being who exercised creative control over the work. When an AI system produces the expressive elements of a work, there is no human author of those elements. There is no copyright. There is nothing to register, license, or enforce.
The report addressed every major argument for extending copyright to AI-generated output and declined each one. The AI cannot be an author because the Act's terms do not reach non-human entities. The user who prompted the AI is not the author of what the AI expressed because they did not make the expressive creative choices. The company that built the AI tool is not the author because they did not author the specific output.
No human authorship over the expressive elements: no copyright.
The mistake most creators make when they read this
The natural response to "AI cannot be an author" is: but I was involved. I opened the interface, I typed the prompt, I made choices. Therefore it's mine.
That reasoning is wrong about what matters.
The question copyright asks is not whether a human was involved. A human is always involved: someone built the account, paid the subscription, typed the instructions. The question is whether the human exercised sufficient creative authorship over the expressive elements of the output.
Prompting is not authorship of the output. When you type a prompt, you are giving the AI instructions. The AI translates those instructions into expression: the specific melody, the specific brushstrokes, the specific word choices, the specific composition. Those expressive choices are the AI's process, not yours. The Copyright Office has addressed this directly: providing text prompts to an AI generally does not constitute sufficient creative authorship over the resulting output to support copyright registration.
The effort you put into refining your prompts is real. The time you spent is real. The creative vision behind your instructions is real. None of that converts the AI's expression into your copyright.
What does count as meaningful human authorship
The copyright analysis turns on whether the human exercised creative control over the expressive elements of the work itself, not over the instructions that produced it.
Four patterns that can cross the line:
Selecting and arranging AI-generated elements. A human who generates many AI outputs and then makes significant creative choices about which to use, how to combine them, and how to arrange them may author the selection and arrangement. This is analogous to a photographer: the camera produces the image, but the photographer's choices of subject, framing, light, and timing are copyrightable creative expression. Scrolling through outputs and clicking the one that looks best is not enough. The selection choices must themselves reflect independent creative judgment.
Significantly modifying AI output. If you generate an AI image and then spend substantial time painting over it, restructuring it, adding elements through your own creative hand, the human-authored modifications may be protectable. The AI-generated base layer is not. The human modifications on top of it are evaluated on their own.
Using AI as one tool within a larger human-authored work. A screenwriter who authors a full script using traditional creative work, then uses an AI tool to generate a single background image for one scene, has human-authored the script. The script is protected. The AI-generated image in that one frame is not.
AI-assisted writing with substantive human creative decisions throughout. If you used an AI tool as a drafting assistant while making the structural, argumentative, and stylistic choices yourself throughout, the work as a whole may reflect sufficient human authorship. The analysis is fact-specific and depends on the actual creative process, not the tools present.
What does not cross the line: selecting a style in a dropdown, choosing between two generated options, adding your name to the output, publishing it, or selling it.
What this means for your work specifically
Musicians. If you used a tool like Suno, Udio, or a similar AI music generator by typing a prompt, you do not have a copyright in the resulting sound recording. The melody, the arrangement, the production, the vocal performance generated by the AI: none of it is yours to register or enforce. If you wrote the lyrics yourself and the AI generated the backing track, your lyrics may be registrable as a literary work. The sound recording of the AI-generated music is not. A song is two copyrights under U.S. law: the composition and the sound recording. AI generation of either one leaves that portion unprotected.
Visual artists. If you used Midjourney, DALL-E, Stable Diffusion, or a similar tool to produce an image by prompting, the output is not registrable. The Copyright Office's Review Board reached this conclusion specifically in the Théâtre D'opéra Spatial case involving AI-generated artwork submitted for registration. If you used AI to generate a raw image and then made significant creative modifications through your own artistic work, the human-authored modifications may be registrable. The AI-generated base is not.
Video creators. AI-generated B-roll, AI-generated voiceover, AI-generated music within a video: each of these elements is unprotectable as AI output. The human-authored script, direction, performance, and creative selection and arrangement of those elements are registrable. A hybrid production results in hybrid copyright: portions protected, portions not. This matters when you try to enforce it.
Writers and course creators. If an AI generated the substantive text of your course, articles, or written content, that text is not protectable. If you used AI as a writing tool while making the structural, argumentative, and stylistic decisions yourself throughout the process, the work as a whole may reflect sufficient human authorship for registration. The analysis depends on what you actually did, not what tool you had open.
How to register AI-assisted work correctly
If your work contains AI-generated elements alongside human-authored elements, you can register the human-authored portions. The Copyright Office requires disclosure when a work contains AI-generated material.
Attempting to register AI-generated work as purely human-authored is a false representation in a federal filing. That is a problem that extends well beyond a refused registration.
For AI-assisted work:
- Disclose the AI-generated portions in your registration application
- Claim copyright only in the human-authored portions
- Describe what the human author contributed
The Copyright Office published specific registration guidance for works containing AI-generated content. That guidance is available at copyright.gov/ai. The USCO Registration Guidance for Works Containing AI-Generated Materials addresses how to complete the application, what to disclose, and how to describe the scope of the human authorship claim.
The two risks of not knowing where you stand
Not understanding your copyright position creates two distinct problems.
The first is the enforcement gap. If you publish AI-generated content believing it is protected, you may have no remedy when someone takes it. No copyright registration means no access to statutory damages. No statutory damages means no practical litigation path for most creators. The infringement happens. You cannot stop it.
The second is the upstream exposure. Over 70 copyright infringement lawsuits have been filed against AI companies in the United States, with additional litigation in the UK and EU. The central question in many of those cases: whether training AI models on copyrighted work constitutes fair use. That question is not settled as of 2026. Courts are deciding it now. Publishing and monetizing AI-generated work carries legal uncertainty about the provenance of that work's training data. Most creators are not tracking this. They should be.
Knowing your position does not eliminate either risk. It tells you what you are working with.
Frequently asked questions
Can I copyright AI-generated art?
No. In the United States, artwork generated solely by an AI system is not eligible for copyright registration. The Copyright Office and the federal courts have confirmed this consistently. If you made significant human creative modifications to AI-generated artwork, the human modifications may be registrable. The AI-generated portions are not, regardless of the modifications.
Does it matter which AI tool I used?
No. The copyright analysis does not turn on the specific tool. It turns on the degree of human creative control exercised over the expressive elements of the output. Midjourney, DALL-E, Suno, ChatGPT, Stable Diffusion: the result is the same under current U.S. law. The tool is not the variable. The human's creative role in the expressive output is.
Can the AI company own the copyright to what their tool generates?
No. Under current U.S. law, there is no copyright in solely AI-generated output for anyone. Not for the user who prompted it. Not for the company that built the tool. Not for the AI system itself. There is no owner because there is no protectable copyright to own.
What if I spent hours crafting my prompts?
Prompting is instruction. The effort and creative thought that goes into a prompt does not transfer into copyright over what the AI produces in response. The Copyright Office addressed this directly in its 2025 guidance. Think of it as directing a performance: a film director makes creative decisions that shape the final work, but the director does not hold the copyright in the actors' performances. The expression belongs to the performer. In AI generation, the expression belongs to the AI's process, not the human's instructions.
Is this the same in Canada?
Canadian copyright law also requires human authorship, and AI cannot hold copyright under the Copyright Act of Canada. The precise framework and relevant judicial decisions differ from the U.S. The analysis of AI-assisted work under Canadian law has not been litigated to the same depth as in the United States. If you are operating in both jurisdictions or producing work primarily for Canadian markets, consult with an attorney licensed in the relevant Canadian jurisdiction. StarGuard Law is licensed in Quebec and Ontario as well as California.
What if AI copyright law changes?
Legislative efforts are active. The Generative AI Copyright Disclosure Act was introduced in the U.S. Congress. State-level proposals have emerged in California and elsewhere. As of May 2026, the federal rule remains: human authorship is required for copyright protection. If the law changes materially, this article will be updated. The USCO Part 3 report on generative AI training, released in pre-publication form in May 2025, is expected to influence future policy and litigation. Creators should monitor these developments.
This article is for general information only — not legal advice.