Copyright and Artificial Intelligence: Ownership, Training Data, and Generated Works

Copyright and Artificial Intelligence: Ownership, Training Data, and Generated Works

Artificial intelligence has introduced one of the most difficult questions in modern copyright law:

Who owns the copyright in something created with artificial intelligence?

The question sounds simple, but it actually contains several different legal questions.

Who owns the material used to train an AI system?

Can copyrighted books, photographs, music, software, and other works legally be used as training data?

Does an AI system itself become an author?

If a person enters a prompt and an AI system generates an image, does the person own the copyright?

What happens when a human substantially edits or rearranges an AI-generated result?

And what happens when an AI-generated output resembles an existing copyrighted work?

These questions do not have one universal answer.

Under current U.S. copyright law, the most important principle is that copyright protects human authorship. Artificial intelligence can be used as a tool in the creative process, but the mere fact that a person instructed an AI system to produce something does not automatically mean that the resulting material is protected by copyright.

At the same time, human creativity incorporated into an AI-assisted work can potentially receive copyright protection.

The U.S. Copyright Office has addressed these issues in a series of reports on copyright and artificial intelligence. Its 2025 report on copyrightability concluded that existing copyright principles are generally capable of addressing AI-generated material without creating a new copyright category for purely machine-generated works.

The result is a developing legal framework in which the key question is increasingly not simply “Was AI used?”, but rather:

“What did the human contribute, and what exactly is being claimed as the copyrighted work?”

---

1. Why Artificial Intelligence Creates a Copyright Problem

Traditional copyright law assumes a human creator.

A novelist writes a novel.

A photographer composes and captures a photograph.

A painter creates a painting.

A musician writes a composition.

A programmer writes software.

Generative artificial intelligence complicates this model because a person can now provide instructions to a machine and receive sophisticated text, images, music, video, software code, or other material in response.

The resulting work may look highly creative.

But appearance alone does not answer the legal question of authorship.

Copyright law asks whether the material contains sufficient original human expression to qualify for protection.

This is why AI has not simply created a new question about technology.

It has revived a fundamental question at the heart of copyright law:

What does it mean to be an author?

---

2. Copyright Generally Requires Human Authorship

Copyright protects original works of authorship.

The Copyright Office has repeatedly emphasized the importance of human creativity in determining whether AI-assisted material can receive copyright protection.

The basic principle is straightforward:

A machine is not treated as the legal author of a copyrightable work merely because it generated the material.

The Copyright Office’s 2025 Copyright and Artificial Intelligence, Part 2: Copyrightability report concluded that existing law should continue to apply the human-authorship requirement and that there is no need for a separate copyright system for purely AI-generated material.

This does not mean that every work involving AI is uncopyrightable.

It means that the presence of AI does not substitute for human authorship.

---

3. AI as a Tool Versus AI as the Claimed Author

The distinction becomes easier to understand if AI is treated as a creative tool.

A human might use:

• a camera;
• image-editing software;
• a word processor;
• digital painting software;
• music-production software;
• or an AI system.

The fact that technology is involved does not automatically eliminate copyright protection.

The legal question is whether the human exercised sufficient creative control over the protected expression.

A person who uses AI as part of a larger creative process may therefore have copyright in portions of the resulting work.

But a person who simply asks an AI system to create an image and exercises no meaningful creative control over the resulting expressive elements may not own a copyright in the AI-generated expression itself.

---

4. A Prompt Does Not Automatically Create Copyright

One of the most common misconceptions about generative AI is:

“I wrote the prompt, therefore I own the copyright in the output.”

That conclusion does not necessarily follow.

A prompt can certainly contain original human expression.

The prompt itself may potentially qualify for copyright protection if it contains sufficient original authorship.

But the copyrightability of the prompt is a separate question from the copyrightability of the AI-generated output.

Suppose someone writes:

“Create a futuristic city at sunset.”

The sentence may or may not contain enough originality to qualify for copyright protection.

But even if the prompt itself were copyrightable, that would not automatically make the resulting AI-generated image the copyrightable work of the person who wrote it.

The output must be analyzed separately.

---

5. Human Creativity Can Exist Alongside AI Generation

The more sophisticated situation occurs when a human uses AI as one component of a larger creative process.

Imagine that an artist:

1. develops a detailed artistic concept;
2. generates numerous AI images;
3. selects particular elements;
4. combines elements from multiple generations;
5. manually redraws substantial portions;
6. changes lighting, composition, perspective, and character design;
7. adds original hand-created elements;
8. edits the final image extensively; and
9. creates a final composition reflecting the artist’s own creative choices.

The resulting work may contain both:

• material generated by AI; and
• original human-authored material.

Copyright analysis can then focus on the human-authored contributions.

The important point is that the copyright does not necessarily cover every element merely because the human participated in creating the overall work.

---

6. Copyright May Protect the Human Contributions

Consider a hypothetical novel created with AI assistance.

An author uses an AI system to generate possible descriptions of a fictional city.

The author then:

• selects particular ideas;
• rewrites the descriptions;
• develops characters;
• creates the plot;
• reorganizes scenes;
• writes original dialogue;
• adds original prose;
• and substantially edits the final manuscript.

The resulting novel may contain significant human-authored expression.

The copyright may therefore protect the author’s original contributions even though AI was involved in the creative process.

This is fundamentally different from saying that the AI itself owns the work.

---

7. Copyright Can Exist in a Larger Work Containing Unprotected AI Material

A work does not necessarily have to be entirely human-created to contain copyrightable material.

Imagine a graphic novel containing:

• AI-generated background scenery;
• human-written dialogue;
• human-created character designs;
• human-written narration;
• and human-composed page layouts.

The AI-generated portions may not independently qualify for copyright protection.

But the human-authored elements and potentially the original selection, coordination, or arrangement of those elements may be protected if they satisfy the applicable standards.

The Copyright Office’s analysis emphasizes this distinction between the overall work and the particular elements for which human authorship can be identified.

---

8. Selection and Arrangement Can Matter

Copyright law can protect sufficiently original selection and arrangement even when individual underlying elements are not themselves protected.

This principle becomes particularly relevant to AI-generated material.

Suppose a designer generates 500 AI images.

The designer then selects 20 images and arranges them into a carefully designed digital exhibition with:

• a particular sequence;
• original captions;
• original textual commentary;
• thematic organization;
• and a distinctive visual arrangement.

The individual AI-generated images might not all be protected as the designer’s original works.

But the human-created selection and arrangement may potentially contain copyrightable expression.

The precise scope of protection depends on the degree of human creativity involved.

---

9. AI-Assisted Works Are Not Automatically “All or Nothing”

This is one of the most important ideas in AI copyright law.

The legal analysis does not necessarily have to produce one of two conclusions:

“Everything is copyrighted.”

or

“Nothing is copyrighted.”

A work can contain different layers of authorship.

For example:

• AI-generated material may be unprotected;
• human-written text may be protected;
• human-created artwork may be protected;
• original arrangement may be protected;
• and factual or public-domain material may remain unprotected.

The resulting copyright may therefore cover only the human-authored portions.

This is sometimes described as protectability of the human contributions.

---

10. The Copyright Office and AI-Generated Works

The U.S. Copyright Office has taken the position that existing copyright law is capable of addressing AI-generated works.

Its Part 2 report on copyrightability examined the human-authorship requirement, the role of prompts, human modification, selection and arrangement, and other forms of human contribution.

The Office’s approach can be summarized conceptually as follows:

AI assistance does not automatically destroy copyrightability, but AI generation does not automatically create copyright either.

The decisive issue is the nature and extent of human authorship.

This approach avoids creating a completely separate category of “AI copyright.”

Instead, it applies traditional copyright principles to a new technological environment.

---

11. What About AI-Generated Text?

The same principle applies to text.

Suppose a person enters:

“Write a 2,000-word article about the history of the Roman Empire.”

The AI produces the article.

If the person merely accepts the output without meaningful creative modification, the purely AI-generated expression may not be protected by copyright merely because the person supplied the instruction.

But suppose the person:

• develops the argument;
• writes substantial original passages;
• restructures the article;
• selects and organizes material;
• rewrites the AI-generated text extensively;
• adds original analysis;
• and creates an original final composition.

The resulting work may contain substantial human-authored expression.

The copyright analysis therefore depends upon the actual creative process rather than merely the presence of an AI tool.

---

12. What About AI-Generated Images?

AI-generated visual art raises the same issue.

Imagine someone enters:

“Create a Renaissance-style portrait of a woman standing beside a window.”

The AI generates an image.

If the person simply accepts the generated image, the human contribution may be insufficient to establish copyright in the AI-generated expressive elements.

Now consider a different process.

The artist:

• creates an elaborate visual concept;
• generates hundreds of alternatives;
• chooses particular compositions;
• modifies the composition manually;
• paints portions of the image;
• changes the facial features;
• adds original clothing details;
• creates the background;
• adjusts lighting;
• and makes extensive creative alterations.

The resulting work contains substantial human contributions.

Those contributions may potentially be protected.

---

13. What About AI-Generated Music?

Music creates similar questions.

Suppose an AI system generates an entire musical composition after receiving a simple instruction.

The resulting composition is not automatically protected merely because a human requested it.

But a human musician might:

• write original lyrics;
• compose portions of the melody;
• arrange the instrumentation;
• edit the structure;
• perform the music;
• modify the recording;
• and combine AI-generated material with original human-authored material.

Different layers of copyright may then exist.

The musical composition, lyrics, and sound recording can involve distinct rights.

AI therefore makes the ownership analysis more complicated, not necessarily less structured.

---

14. AI-Generated Software Code

Software introduces another important category.

A developer may ask an AI system to generate code.

The developer then incorporates portions of that code into a larger program.

The legal analysis can involve questions such as:

• Who wrote the code?
• How much of the code was generated by AI?
• What human modifications were made?
• Was the code copied from existing material?
• Does the output contain third-party copyrighted code?
• What license terms apply to any underlying source material?
• Does the final program contain sufficient human-authored expression?

The fact that AI produced the initial code does not automatically answer these questions.

Software copyright can be particularly complicated because code also interacts with open-source licenses, contractual restrictions, and other intellectual-property rights.

---

15. Who Owns an AI-Generated Work?

The question “Who owns an AI-generated work?” is therefore sometimes based on a false premise.

If the material is entirely AI-generated and contains no sufficient human authorship, the central issue may not be who owns the copyright.

The more fundamental question may be:

Is there a copyright to own at all?

If no copyright exists in the AI-generated expression, the person who generated it does not necessarily receive an exclusive copyright merely because they operated the AI system.

That is fundamentally different from ordinary human authorship.

---

16. Can an AI System Be the Copyright Author?

Under current U.S. copyright principles, an AI system itself is not treated as a human author capable of owning copyright.

Copyright law is structured around human authorship.

The Copyright Office’s 2025 report likewise rejected the need to create a new copyright regime granting authorship to AI systems.

This means that an AI model does not become a copyright owner simply because it generated a poem, image, song, video, or computer program.

The legal system must instead identify the human contribution, if any.

---

17. What Is AI Training Data?

AI training introduces a different copyright question.

Generative AI systems are developed by exposing models to enormous quantities of data.

Training data can include:

• books;
• articles;
• websites;
• photographs;
• illustrations;
• music;
• videos;
• software code;
• scientific papers;
• and other forms of content.

Some of this material may be protected by copyright.

This creates a fundamental legal question:

Can copyrighted works be copied and used to train an AI model without permission?

There is currently no simple universal answer under U.S. law.

The question is being addressed through litigation, regulatory analysis, licensing arrangements, technological developments, and evolving policy discussions.

---

18. Training Data and Copyright Are Not the Same Question as AI Output

It is important to separate two different issues.

Question One: Training

Was copyrighted material lawfully copied or processed to train an AI model?

Question Two: Output

Does the material generated by the AI system infringe someone’s copyright?

These questions are related but legally distinct.

A model could potentially be trained using lawful sources but later generate infringing material.

Conversely, a dispute concerning training data does not automatically establish that every output produced by the trained model infringes copyright.

AI copyright analysis therefore requires looking at different stages of the technological process.

---

19. Is AI Training Fair Use?

One of the central legal questions in the United States is whether certain uses of copyrighted works for AI training qualify as fair use.

Fair use is a doctrine under § 107 of the Copyright Act that can permit certain unauthorized uses of copyrighted material.

The four statutory factors consider:

1. the purpose and character of the use;
2. the nature of the copyrighted work;
3. the amount and substantiality of the portion used; and
4. the effect of the use upon the potential market for the copyrighted work.

Cornell’s Wex explanation of fair use describes this as a balancing test rather than an automatic rule.

That makes fair use particularly difficult to apply mechanically to AI training.

---

20. Why AI Training Creates Difficult Fair-Use Questions

Traditional fair-use cases involved particular uses of copyrighted works.

AI training can involve enormous datasets containing millions or billions of works.

The technology may copy works during training, transform information into statistical or computational representations, and later generate new material.

This creates difficult questions about:

• the purpose of the copying;
• whether the use is transformative;
• whether commercial AI development changes the analysis;
• whether the original works are expressive;
• how much material was copied;
• whether complete works were copied;
• whether the training process affects markets for original works;
• and whether AI-generated outputs compete with the originals.

The Copyright Office’s Part 3 report specifically examines these questions and emphasizes that the legal analysis depends on the particular facts and uses involved.

---

21. Commercial AI Training

Commercial purpose can be relevant to fair use, but it is not automatically decisive.

An AI company may train a commercially valuable model.

That fact does not automatically mean that training is infringement.

Nor does it automatically establish fair use.

Fair use requires consideration of the statutory factors as a whole.

The commercial nature of a use is one part of the analysis.

Courts must consider the overall character and consequences of the use.

---

22. The Market-Effect Question

One of the most consequential issues concerns the market for copyrighted works.

Imagine that an AI system is trained on millions of books.

The model can then generate text that users employ instead of purchasing books, commissioning writers, or licensing certain content.

Does that affect the market for the original works?

If so, how?

The question can become even more complicated when AI-generated material competes with the market for the original work itself.

The Copyright Office’s analysis recognizes the importance of market effects and the possibility that generative AI may affect existing and emerging markets for copyrighted works.

This is likely to remain one of the most important issues in future copyright litigation.

---

23. Licensing Training Data

One possible solution to the legal uncertainty is licensing.

AI developers can obtain permission to use copyrighted material from copyright owners.

For example, an AI company might negotiate agreements with:

• publishers;
• news organizations;
• stock-image libraries;
• music companies;
• software repositories;
• academic databases;
• or individual creators.

A license can provide a clearer legal foundation for using the material.

This approach also creates a potential economic market in which creators are compensated for making their works available for AI development.

---

24. Opt-Out and Access Restrictions

Another important issue concerns whether copyright owners can restrict AI systems from accessing their works.

Some websites have implemented technical measures intended to discourage automated crawling.

Others have adopted terms of service addressing AI training.

Still others have entered licensing agreements instead of attempting to prohibit AI use entirely.

The effectiveness and legal significance of these mechanisms can vary.

A technical restriction, contractual term, and copyright rule are not necessarily the same thing.

---

25. Training on Public-Domain Works

AI systems can also be trained on works in the public domain.

Public-domain works generally are not protected by copyright.

For example, depending on the particular work and jurisdiction, historical literary works whose copyrights have expired may be available for use without obtaining permission from a copyright owner.

However, public-domain status must be determined carefully.

A modern edition of an old work can contain new copyrightable material even though the underlying historical text is public domain.

The distinction between the original public-domain material and later protected additions remains important.

---

26. Training on Facts

Copyright does not protect facts themselves.

This principle can matter when AI systems process factual information.

For example, a database containing historical facts may include information that is not itself protected by copyright.

But the particular selection, arrangement, expression, or compilation of those facts may potentially be protected.

The distinction is therefore between:

information itself

and

copyrightable expression or arrangement of that information.

AI systems can process facts without necessarily infringing copyright simply because the facts originated in copyrighted works.

But copying expressive text containing those facts can raise different questions.

---

27. AI Output That Resembles Existing Copyrighted Works

Another major problem occurs when AI generates material that resembles an existing work.

Suppose an AI system produces an image that is strikingly similar to a copyrighted illustration.

Or it produces text that reproduces distinctive passages from an existing book.

Or it generates software code that is substantially similar to existing copyrighted code.

The fact that AI produced the output does not automatically eliminate copyright liability.

The output must still be examined under ordinary copyright principles.

The central questions can include:

• Was protected expression copied?
• How substantial is the similarity?
• Is the material original or commonplace?
• Was the output independently generated?
• Did the AI reproduce recognizable protected expression?
• Does an exception such as fair use apply?

AI is not a blanket defense against copyright infringement.

---

28. Style and Copyright

AI has also generated intense debate about artistic style.

Suppose an AI system is instructed:

“Create a painting in the style of a particular living artist.”

Copyright law generally does not give a creator ownership of an abstract artistic style in the same way it protects a particular copyrighted work.

Copyright generally does not protect ideas, concepts, methods, or styles as such.

Cornell’s explanation of copyright similarly emphasizes that copyright protects expression rather than ideas or concepts.

However, this does not mean that anything resembling another artist is automatically lawful.

A particular generated work could potentially reproduce protected elements of an existing work.

Other areas of law, including trademark, right-of-publicity, unfair competition, or other doctrines, may also become relevant depending on the circumstances.

Thus:

Style and copyrighted expression are not necessarily the same thing.

---

29. AI and Derivative Works

AI can also be used to create adaptations of existing works.

For example:

• translating a novel;
• turning a novel into a screenplay;
• transforming a photograph into a new illustration;
• creating a remix of music;
• adapting a character into a video game;
• or modifying an existing visual work.

These activities can implicate the copyright owner’s exclusive right to prepare derivative works.

Cornell’s explanation of derivative works notes that a derivative work is based on a preexisting copyrighted work and that copyright in the new work does not eliminate the copyright owner’s rights in the original.

Using AI does not erase those underlying rights.

---

30. AI and Copyright Registration

AI-generated material also creates questions during copyright registration.

An applicant should not simply assume that everything contained in an AI-assisted work is human-authored.

The Copyright Office has instructed applicants to disclose AI-generated material when relevant and to identify the human-authored contributions being claimed.

The underlying principle is transparency.

If a work contains both:

• human-created material; and
• AI-generated material,

the applicant should identify the material for which copyright protection is actually being claimed.

This reflects a broader principle:

Copyright registration should accurately describe the authorship of the work.

---

31. Why Human Editing Matters

Human editing can be legally significant because it can introduce original expression.

Imagine that an AI produces a 1,000-word article.

A human editor then:

• rewrites 600 words;
• changes the structure;
• adds original paragraphs;
• creates original examples;
• removes material;
• reorganizes the argument;
• and substantially changes the language.

The final article may contain significant human-authored expression.

The fact that AI produced the initial draft does not necessarily prevent copyright protection for those human contributions.

The legal question is the nature of the actual human expression.

---

32. Human Selection Can Matter Too

Human creativity does not always take the form of writing or drawing.

Selection can itself involve authorship.

Imagine an AI system generates 10,000 photographs.

A human curator chooses 100 of them and organizes them into a carefully constructed exhibition.

If the selection and arrangement reflect sufficient original human judgment, the resulting compilation may contain copyrightable authorship.

Again, the copyright does not necessarily extend to every underlying AI-generated image.

It may extend to the human-authored selection and arrangement.

---

33. AI Does Not Eliminate the Need to Track Sources

Creators using AI should keep records of their creative process.

Useful records may include:

• prompts;
• drafts;
• source materials;
• versions;
• human edits;
• manually created elements;
• selected outputs;
• discarded outputs;
• and final compositions.

Why?

Because if copyright ownership is later challenged, evidence of the human creative process may become important.

For a professional creator, maintaining such records can therefore become part of intellectual-property management.

---

34. AI and Copyright Infringement Risk for Creators

Creators who use AI should not assume that the output is automatically safe simply because the AI generated it.

Potential risks include:

• output that reproduces protected expression;
• output containing recognizable characters;
• copied passages of text;
• software code derived from protected code;
• unauthorized use of copyrighted images;
• and material generated from sources that the user did not have permission to exploit.

The responsibility for using the final output can depend on the circumstances and applicable agreements.

A creator should therefore review important AI-generated material before commercial publication.

---

35. AI and Licensing Agreements

AI has also changed copyright licensing itself.

Contracts may now address questions such as:

• May the licensed work be used to train AI?
• May the licensee use AI to create derivative material?
• May the work be uploaded into an AI system?
• May confidential material be used as an AI prompt?
• Does the license permit machine learning?
• Who owns AI-assisted output?
• Who is responsible if the output infringes another person’s copyright?

These questions are increasingly appearing in publishing, employment, software, advertising, media, and creative-industry contracts.

A traditional copyright license that says simply “for all permitted uses” may not adequately answer every AI-specific question.

---

36. Employment and AI-Generated Works

AI also complicates questions about works created within employment.

Suppose an employee uses an AI system to create marketing materials for an employer.

The legal analysis may involve several separate issues:

• copyrightability of the AI-generated material;
• human contributions;
• employment agreements;
• work-made-for-hire principles;
• ownership of human-authored contributions;
• company policies;
• and contractual restrictions concerning AI tools.

The mere fact that the employee created the material at work does not answer every copyright question.

The AI component must still be analyzed.

---

37. Confidentiality and AI Training

Copyright is not the only legal concern when copyrighted or proprietary material is entered into an AI system.

A company might possess:

• trade secrets;
• confidential business information;
• unpublished manuscripts;
• customer information;
• proprietary research;
• or private legal documents.

Uploading such material into an AI service may create contractual, confidentiality, privacy, or trade-secret issues even if no copyright infringement occurs.

Thus, an organization should not ask only:

“Do we own the copyright?”

It should also ask:

“Are we legally permitted to disclose this material to the AI system?”

Copyright is only one component of the broader legal analysis.

---

38. International Differences

AI copyright law is not identical throughout the world.

Different countries have different rules concerning:

• text and data mining;
• copyright exceptions;
• authorship;
• AI-generated works;
• training data;
• licensing;
• and database rights.

The U.S. approach should therefore not be treated as a universal global rule.

The U.S. Copyright Office’s international analysis has noted, for example, that many European Union member states take the position that purely AI-generated material cannot receive copyright because only a natural person can be an author, while national approaches can differ in other respects.

For works distributed internationally, jurisdiction can therefore matter substantially.

---

39. The Difference Between Current Law and Future Law

AI copyright law is still developing.

Courts are deciding cases involving AI companies, creators, publishers, artists, programmers, and other rights holders.

Legislatures are considering new laws.

Regulators and copyright offices are studying the technology.

The U.S. Copyright Office has issued multiple reports addressing AI and copyright, including separate work on copyrightability and training.

Congress is also considering legislation concerning AI and copyright-related issues. The Copyright Office’s current legislative-development materials list several AI-related bills introduced in the 119th Congress.

Consequently, anyone dealing with AI-generated content should be cautious about treating today’s legal position as permanently settled.

---

40. A Practical Example: An AI-Assisted Novel

Consider an author named Elena.

Elena asks an AI system to suggest possible plot ideas.

She then develops one of those ideas herself.

She writes the characters, dialogue, descriptions, and narrative.

She uses AI occasionally to suggest alternative wording.

She rejects most suggestions and rewrites the rest.

Finally, she produces a 90,000-word novel.

In this situation, AI has functioned as an assistive tool within a predominantly human creative process.

The existence of AI assistance does not automatically eliminate copyright protection for Elena’s human-authored expression.

Now consider a different situation.

Elena asks an AI system:

“Write a complete 90,000-word novel in my preferred style.”

The system generates the entire manuscript.

Elena makes only minor spelling corrections.

The copyright analysis is very different.

The central question would be how much protectable human authorship actually exists in the final manuscript.

The two situations involve the same technology but very different creative contributions.

---

41. A Practical Example: AI-Generated Artwork

Now imagine an artist named Daniel.

Daniel asks an AI system to create a landscape.

The system generates the entire image.

Daniel selects one result and publishes it unchanged.

The human contribution may be insufficient to establish copyright in the AI-generated expressive content.

But suppose Daniel instead uses AI as a starting point.

He:

• selects a composition;
• manually redraws the characters;
• paints the foreground;
• creates an original sky;
• changes the architecture;
• adds original textures;
• modifies lighting;
• and substantially alters the final composition.

The final image may contain significant human-authored elements.

The legal analysis therefore changes because the creative contribution has changed.

---

42. A Practical Example: Training Data

Now consider an AI company that wants to train a commercial model.

It collects:

• books;
• newspaper articles;
• photographs;
• illustrations;
• music;
• and software code.

Some of these materials are copyrighted.

The company copies the works into a training dataset.

Several legal questions immediately arise.

Did the company have permission?

Were the works lawfully accessed?

Does the copying qualify as fair use?

Does the training process reproduce protected expression?

Does the model retain or reproduce expressive material?

Does the resulting AI system compete with markets for the original works?

Do licensing agreements or terms of service affect the analysis?

These questions cannot be answered merely by saying:

“AI training is transformative.”

Nor can they be answered merely by saying:

“The works were copyrighted, therefore training is infringement.”

The actual legal analysis requires consideration of the relevant facts and applicable law.

---

43. What Creators Should Know About AI and Copyright

Creators using AI should consider several practical safeguards.

Keep records of human contributions

Save drafts, edits, prompts, sketches, source materials, and intermediate versions.

Know the AI service’s terms

The contractual terms of an AI platform may address ownership, licensing, data use, training, confidentiality, and output.

Do not assume every output is copyrightable

The presence of human involvement does not automatically make every AI-generated element protected.

Review important outputs

AI-generated material can contain unexpected similarities, factual errors, or third-party material.

Be careful with copyrighted source material

Uploading copyrighted works to an AI system can raise questions independent of the copyright status of the output.

Consider registration strategically

If a commercially important work contains AI-generated material, identify and document the human-authored portions before seeking copyright registration.

---

44. What Businesses Should Know About AI and Copyright

Businesses should consider AI copyright issues when adopting generative AI tools.

Important questions include:

• What data may employees upload?
• Who owns the resulting work?
• Can the business commercially exploit the output?
• Does the AI provider retain rights to submitted data?
• Is customer information being processed?
• Could confidential material enter the training process?
• Does the output contain third-party material?
• Are employees required to disclose AI use?
• Are there contractual restrictions on AI-generated content?
• How will ownership be documented?

AI policy should therefore be treated as part of broader intellectual-property governance.

---

45. AI and the Future of Copyright

Artificial intelligence is unlikely to eliminate copyright law.

It is more likely to force copyright law to answer its oldest questions with greater precision.

What is authorship?

What is originality?

What counts as copying?

What constitutes transformation?

What does it mean to create a derivative work?

What is a lawful use of another person’s expression?

What is the boundary between an idea and protected expression?

AI does not make these questions obsolete.

It makes them unavoidable.

The technology has created a world in which a human can participate in a creative process without personally producing every expressive element.

Copyright law must therefore distinguish between human creativity assisted by machines and expression generated without sufficient human authorship.

---

Key Takeaways

The most important principles concerning copyright and artificial intelligence are:

• Copyright law generally requires human authorship.
• An AI system is not treated as the copyright author merely because it generated a work.
• The use of AI does not automatically prevent copyright protection.
• Human-authored portions of an AI-assisted work may be protected.
• Human selection, arrangement, editing, and modification can be legally significant.
• A simple prompt does not automatically give the user copyright in the resulting output.
• A completely AI-generated work may contain material that is not protected by copyright.
• Copyright can potentially protect human contributions to a larger work containing AI-generated material.
• AI training and AI-generated output raise separate copyright questions.
• The use of copyrighted works as AI training data can raise fair-use questions.
• Fair use requires consideration of the statutory factors and does not automatically apply merely because training is technologically transformative.
• AI-generated material can still infringe copyright if it reproduces protected expression.
• Copyright does not generally protect abstract artistic styles, although particular expressive works may be protected and other legal doctrines may apply.
• Licensing is one possible way for AI developers to obtain lawful access to copyrighted training material.
• AI contracts and platform terms can affect ownership, licensing, confidentiality, and data use.
• Copyright registration should accurately identify human-authored contributions.
• The law is still developing, and AI-related copyright rules may change through legislation and judicial decisions.

---

Frequently Asked Questions

Can I copyright something created by ChatGPT or another AI system?

Not automatically. Under current U.S. copyright principles, copyright protection depends upon human authorship. A purely AI-generated output may not receive copyright protection merely because a person entered the prompt.

Does writing the prompt make me the copyright owner?

Not necessarily. A prompt may itself contain copyrightable human expression, but that does not automatically make the AI-generated output a copyrightable work of the person who wrote the prompt.

Can AI-assisted work be copyrighted?

Yes, potentially. Human-authored expression incorporated into an AI-assisted work can receive copyright protection if it satisfies the requirements of copyright law.

Can I copyright an AI-generated image that I edited?

Potentially. The important question is what original human expression was added through the editing process. The copyright may protect the human-authored contributions rather than the purely AI-generated material.

Can an AI own copyright?

Under current U.S. copyright principles, an AI system is not treated as a human author capable of owning copyright.

Is AI training copyright infringement?

Not automatically. The legality of using copyrighted works for AI training can depend on authorization, fair use, the nature of the copying, the purpose of the use, market effects, and other circumstances.

Is AI training fair use?

It can be, but there is no blanket rule that all AI training is fair use. The Copyright Act’s four-factor fair-use analysis must be applied to the relevant circumstances.

Can an AI company train on copyrighted books without permission?

The answer depends on the circumstances and applicable law. Whether unauthorized training is lawful is an active area of litigation, legal analysis, licensing negotiations, and policy development.

If AI produces something similar to a copyrighted work, is that infringement?

It can be, depending on the nature and extent of the similarity and other relevant circumstances. AI generation does not automatically provide immunity from copyright infringement.

Does copyright protect an artist’s style?

Copyright generally does not grant exclusive ownership of an abstract style. However, copying protected expression from particular works can raise copyright issues, and other areas of law may sometimes apply.

Can I use AI to create a derivative work?

AI does not eliminate the copyright owner’s exclusive rights. If the process uses protected expression from an existing copyrighted work, permission may be required unless an applicable exception applies.

Should I disclose AI use when registering a copyright?

Where AI-generated material is included in a work being registered, applicants should accurately identify the human-authored portions and comply with the Copyright Office’s current registration requirements.

Is AI copyright law settled?

No. Many basic principles are established, particularly the importance of human authorship, but questions concerning training data, fair use, outputs, licensing, and market effects remain actively contested and may continue to develop.

---

Conclusion

Artificial intelligence has not made copyright law irrelevant.

It has made the question of authorship more important than ever.

The central distinction is between using AI as a tool within a human creative process and claiming copyright in material that was generated without sufficient human authorship.

A human who writes, edits, selects, arranges, modifies, illustrates, composes, or otherwise contributes original expression may have copyrightable interests in those contributions even when artificial intelligence played a substantial role in the creative process.

At the same time, a person cannot necessarily acquire copyright in purely machine-generated expression simply by operating an AI system or writing a prompt.

The second major issue concerns the other end of the process: the data used to create AI systems in the first place.

Training sophisticated models can involve the copying and processing of enormous quantities of copyrighted material. Whether particular training practices are lawful may depend on permission, licensing, fair use, the nature of the copying, and the effects on existing and potential markets. The U.S. Copyright Office’s separate report on generative-AI training demonstrates just how significant and unresolved these questions remain.

The emerging legal framework can therefore be summarized in three propositions:

AI does not automatically become an author.

AI assistance does not automatically destroy human copyright.

And AI training does not automatically make the use of copyrighted material either lawful or unlawful.

The decisive questions remain human authorship, protected expression, authorization, copying, transformation, and the particular facts of each case.

Artificial intelligence may be a new technology, but the underlying legal problem is an old one:

Who created the expression, what expression is protected, and who has the legal right to use it?