A person sits at a computer and describes an image that does not exist.
Seconds later, artificial intelligence creates it.
The person changes the instructions.
The system creates another version.
More instructions follow. The composition changes. Objects move. The lighting changes. Eventually, an image appears that looks exactly the way the person wanted.
Who created it?
The person who imagined it?
The company that built the artificial-intelligence system?
The people whose works were used to train the system?
Or no copyright owner at all?
Generative artificial intelligence has made it possible for almost anyone to produce sophisticated images, music, writing, video and computer code with remarkably little traditional creative labor.
Copyright law now has to determine where the machine ends and the human author begins.
That distinction matters because under current U.S. copyright law, human authorship remains fundamental.
Copyright Law Was Built Around Human Creators
The Constitution gives Congress power to protect the writings of authors.
Congress built a copyright system around creative works produced by people.
For centuries, that assumption barely needed explanation.
Books had authors.
Paintings had artists.
Songs had composers.
Photographs had photographers.
Then machines became increasingly involved in the creative process.
A camera creates an image mechanically, but a photographer makes creative decisions about composition, subject, timing and other expressive elements.
A word processor assists a novelist without becoming the author.
Editing software helps a filmmaker manipulate images without acquiring copyright itself.
Generative AI creates a harder problem because the machine can produce expressive material that the user never specifically designed.
The user may describe a desired result.
The system determines many of the details.
That gap between instruction and output has become central to the copyright debate.
The Copyright Office Still Requires Human Authorship
The U.S. Copyright Office has repeatedly maintained that copyright protects works of human authorship.
In its 2025 report addressing the copyrightability of generative-AI outputs, the Office concluded that existing copyright law is capable of handling AI-generated material and that purely AI-generated material is not copyrightable. It also concluded that using AI as a tool does not prevent copyright protection for human-created portions of a work.
That creates an important distinction.
Using artificial intelligence does not automatically eliminate copyright.
The question is what the human actually contributed.
A person can use AI during a larger creative process and still own copyright in the person's original expression.
But simply requesting an output does not necessarily make the resulting machine-generated expression copyrightable.
A Prompt Is Not Necessarily the Same Thing as Creating the Result
Suppose someone types:
“Create a painting of a lighthouse during a storm.”
The AI generates the image.
The user did not determine the exact shape of every wave.
The user did not decide precisely where each cloud would appear.
The user did not choose every color or brush-like mark.
The system supplied those expressive details.
Now make the prompt much longer.
The user describes the lighthouse, weather, viewpoint, lighting, ocean, composition and atmosphere in extraordinary detail.
Has the user now become the author of the image?
Under the Copyright Office's current approach, prompts generally function as instructions rather than giving the user sufficient control over the particular expressive elements generated by the system.
That does not mean prompts have no creative value.
A prompt itself may contain human-authored expression.
But copyright in the words of a prompt is a different question from copyright in the image the system produces in response.
The Monkey-Selfie Dispute Foreshadowed the Human-Authorship Problem
Long before generative AI became widespread, an unusual photograph helped illustrate the problem of nonhuman creation.
A macaque named Naruto took photographs of itself using wildlife photographer David Slater's camera equipment.
The photographs became famous.
Litigation followed over whether the monkey could hold copyright.
In Naruto v. Slater, the Ninth Circuit held that the monkey lacked statutory standing under the Copyright Act. The dispute became one of the best-known modern examples of copyright law confronting a work created without ordinary human authorship.
The case was not about artificial intelligence.
But it exposed a question that would soon become much more important:
What happens when something that looks like a creative work is produced without a human being acting as its author?
Generative AI has transformed that unusual question into an everyday one.
A Computer Scientist Tried to Register AI-Created Artwork
The human-authorship issue moved directly into the AI era through litigation involving computer scientist Stephen Thaler.
Thaler sought copyright registration for an artwork titled A Recent Entrance to Paradise.
He identified an artificial-intelligence system as the work's author.
The Copyright Office refused registration because the work lacked human authorship.
Thaler challenged that decision in federal court.
In 2023, the U.S. District Court for the District of Columbia upheld the Copyright Office's decision, concluding that human authorship is an essential requirement of copyright.
The D.C. Circuit affirmed in 2025, holding that the Copyright Act requires all eligible works to be authored in the first instance by a human being.
The decision answered an important question.
An AI system cannot simply be named as the copyright author under existing federal law.
But it did not resolve every harder question about humans working with AI.
The Real Difficulty Begins When Humans and AI Work Together
Most AI-assisted works are not as simple as asking whether a machine can own copyright.
Imagine a graphic designer creates an advertisement.
The designer writes the headline.
The designer creates the layout.
The designer chooses the typography.
An AI system generates an illustration used in the background.
The designer then modifies portions of that image manually.
What is protected?
Potentially, the human-authored elements.
The text may qualify.
The arrangement may qualify.
The human modifications may qualify.
The selection and coordination of material may qualify.
But the portions generated entirely by AI may remain outside copyright protection.
A single finished work can therefore contain both copyrightable and uncopyrightable material.
The Comic Book Zarya of the Dawn Became an Early Test
One of the most important early Copyright Office decisions involved Zarya of the Dawn, a comic book created by Kristina Kashtanova.
Kashtanova wrote the text and arranged the material but used Midjourney to generate images.
The Copyright Office initially registered the work without understanding the extent of AI involvement.
After reviewing the circumstances, the Office concluded that Kashtanova was the author of the text and the selection and arrangement of the material, but not the individual images generated by Midjourney.
The Office reasoned that the user did not exercise sufficient control over the specific expressive output produced by the AI system.
The result demonstrated that copyright analysis does not necessarily have to treat an AI-assisted work as all-or-nothing.
Human contributions can be protected even when machine-generated components are not.
Editing AI Output Can Create New Human Authorship
Now imagine the AI produces an image.
The user opens it in editing software.
The user substantially redraws the face.
Changes the background.
Adds original objects.
Alters the composition.
Paints new details.
At some point, human expression has clearly entered the work.
Copyright can protect the human-created modifications even if the underlying AI-generated material itself is not protected.
The Copyright Office's guidance recognizes this possibility.
The analysis depends on what the human contributed rather than simply whether AI appeared somewhere in the creative process.
This makes documentation increasingly important for professional creators.
The final image alone may not reveal which portions came from a human and which came from a machine.
Arrangement Can Be Creative Even When Individual Pieces Are Not
Copyright can also protect creative selection and arrangement.
Imagine someone generates hundreds of AI images.
The individual images may lack copyright protection if they were generated without sufficient human authorship.
But the person carefully selects twenty of them and arranges those images into an original visual narrative.
The arrangement can involve human creativity.
A similar principle already exists outside artificial intelligence.
A compilation can receive copyright protection in the original selection or arrangement of material even when the individual components are not themselves protected.
The protection does not suddenly give the compiler ownership of every underlying item.
It protects the human-authored contribution to the compilation.
Ownership and Copyrightability Are Different Questions
People often ask:
“Who owns this AI image?”
But that can combine several different legal questions.
A service's contract may give a user certain rights to use an output.
The user may possess the digital file.
The platform may agree not to assert certain rights against the user.
Those contractual arrangements do not necessarily mean federal copyright law recognizes copyright in the underlying AI-generated expression.
Possessing a file is not the same as owning copyright.
Having contractual permission to use something is not the same as owning copyright.
And being the person who requested a work is not necessarily the same as being its legal author.
Those distinctions become increasingly important as AI-generated content moves into commercial products.
An Uncopyrightable Image May Be Usable by Others
Suppose an image is generated entirely by AI and contains no copyrightable human authorship.
If copyright does not protect the image, the person who generated it may have much less ability to stop other people from copying that image under copyright law.
That can surprise users.
Someone may spend hours experimenting with prompts until the system produces exactly the desired result.
The result may feel deeply personal.
The user may reasonably describe it as “my image.”
But effort alone does not necessarily create copyright.
Copyright focuses on protectable authorship.
This distinction can matter enormously when AI-generated material becomes commercially valuable.
Companies Still Have Contracts
The absence of copyright does not mean there are no legal rules.
AI platforms operate under terms of service.
Those contracts can govern the relationship between the company and the user.
They can address permitted uses, account rights, responsibility for outputs and other issues.
Businesses can also use confidentiality agreements, employment agreements, licensing arrangements and other contractual protections.
Trademark, patent, trade-secret, publicity and unfair-competition laws can sometimes become relevant depending on the circumstances.
Copyright is important, but it is not the entire legal system governing creative material.
Training AI Creates a Completely Different Copyright Question
The ownership of AI outputs is often confused with another major legal dispute:
Was copyrighted material lawfully used to train the AI system?
Those are separate questions.
A novelist may argue that copies of books were unlawfully used during training.
An AI company may argue that its training practices are protected by fair use or otherwise lawful.
That dispute concerns inputs.
Whether a particular AI output qualifies for copyright protection concerns the output.
The legal answer to one does not automatically determine the other.
Courts have been confronting training-related copyright lawsuits brought by authors, visual artists, news organizations and other rights holders, making this one of the most consequential unresolved areas of AI law.
An AI Output Can Still Infringe Someone Else's Copyright
There is another misconception worth separating.
Suppose an AI-generated image itself is not copyrightable.
That does not necessarily mean the image cannot infringe an existing copyrighted work.
Imagine a system produces material substantially similar to protected expression from an existing work.
The fact that nobody owns copyright in the new AI-generated material would not automatically erase the original copyright owner's rights.
Copyrightability asks:
Can this new work receive copyright protection?
Infringement asks:
Does this new material unlawfully copy protected expression from an existing work?
Those are different inquiries.
The Same Problem Applies to AI-Generated Music
Music makes the issue especially vivid.
A person can ask an AI system to generate a song.
The system may produce melody, harmony, instrumentation and vocals.
What did the user actually author?
Perhaps the user wrote lyrics.
Those human-written lyrics can present a straightforward copyright claim.
Perhaps the user wrote the melody and used AI only to create instrumentation.
Again, there may be substantial human authorship.
But if the person merely describes the kind of song desired and the system determines all of the expressive musical details, the copyright analysis changes.
As AI becomes capable of generating increasingly sophisticated music, the boundary between tool and creator becomes commercially significant.
Software Raises the Same Boundary Problem
Programmers increasingly use AI systems to generate computer code.
Sometimes the programmer writes most of the program and uses AI to suggest individual functions.
Sometimes the system generates substantial blocks of code.
Sometimes the programmer repeatedly modifies and restructures the output.
The same basic question emerges:
What did the human author?
The answer may vary throughout a single software project.
And code creates an additional concern: generated material may resemble code appearing in training sources, potentially raising separate licensing or infringement issues.
AI does not eliminate traditional intellectual-property questions.
It layers new ones on top of them.
Employers Will Have to Think About Human Authorship Too
Businesses routinely rely on copyright ownership.
Advertising agencies own or license creative work.
Software companies depend on copyright in code.
Publishers acquire rights from authors.
Studios rely on enormous portfolios of protected material.
Now imagine a company asks an employee to generate hundreds of AI images for a commercial project.
If the resulting material contains little or no human authorship, traditional assumptions about copyright ownership may not work as expected.
The issue is not necessarily whether the company or employee owns the copyright.
There may be no copyright in the purely AI-generated portions to own.
Companies using generative AI commercially therefore have reason to understand how much human creative work goes into the final product.
Keeping Records of the Creative Process May Become Valuable
Historically, a finished work often spoke for itself.
A painter painted the painting.
A photographer took the photograph.
A writer wrote the manuscript.
With AI-assisted work, the final product may not reveal the creative process.
Two nearly identical images could have very different copyright status.
One might be generated almost entirely by AI.
The other might begin with AI but then undergo extensive human modification.
Documentation can help establish the difference.
Drafts.
Source files.
Editing histories.
Original sketches.
Human-written text.
Layered image files.
Records showing how the work changed.
These materials can help demonstrate where human authorship entered the process.
AI Is Forcing Copyright Law to Separate Ideas From Execution in a New Way
A person may have a brilliant idea for an image.
A castle floating over an ocean at sunset.
That concept can be highly creative.
But copyright traditionally does not protect ideas themselves.
It protects original expression.
Generative AI makes the distinction unusually visible.
The user can supply the idea in a prompt.
The machine can supply much of the final expression.
The person may feel like the creator because the concept originated in the person's mind.
Copyright law asks a more specific question:
Who determined the protectable expressive elements of the finished work?
That question can produce a different answer.
The Tool Does Not Automatically Become the Author
None of this means artists lose copyright merely because technology assists them.
Artists have always used tools.
Cameras.
Printing presses.
Musical instruments.
Editing software.
Digital brushes.
Computer graphics.
Word processors.
Copyright law does not require creation with bare hands.
The Copyright Office has emphasized that AI can function as an assistive tool without preventing copyright protection for the human-authored work.
The difficulty arises when the system does more than assist in executing human expression and instead determines expressive elements itself.
The line can be difficult to draw.
But the existence of a machine somewhere in the workflow is not enough to eliminate copyright.
The Law Is Likely to Matter More as AI Work Becomes Valuable
When AI images were mostly curiosities, ownership could seem theoretical.
That changes when an AI-assisted character becomes the centerpiece of a successful business.
When generated artwork appears in advertising.
When an AI-created song becomes popular.
When software generated with AI becomes commercially important.
When a publisher releases books containing machine-generated material.
The more valuable the output becomes, the more important it is to know what legal rights actually protect it.
A person may discover the answer only when someone else copies the work.
At that point, the distinction between emotional ownership and legal copyright becomes very real.
Artificial Intelligence Has Made Creation Easier Than Ownership
Generative AI can produce in seconds what once required hours, days or weeks of work.
That technological achievement creates a legal paradox.
The easier it becomes to produce sophisticated creative material without direct human execution, the harder it can become to identify the human authorship copyright law requires.
The machine can generate the image.
The song.
The paragraph.
The video.
The computer code.
But under current U.S. copyright law, the machine cannot simply step forward and become the author.
That leaves courts, the Copyright Office and creators with the more difficult task of separating what the human contributed from what the system supplied.
For now, the emerging principle is relatively clear even though its application can be complicated:
Artificial intelligence can be a powerful creative tool, but copyright still begins with human creativity.