Generative artificial intelligence is transforming how people write books, create images, compose music, produce videos and develop software. But behind this technological revolution is a fundamental legal question: can AI companies use copyrighted material to train their models without permission, and who owns the content those systems generate?
Authors, publishers, photographers, musicians, film studios and software developers are challenging AI companies in courts around the world. At the same time, technology companies argue that training on existing material is necessary to develop useful AI systems and may qualify as fair use or another copyright exception.
The outcome of these disputes could reshape the economics of creative industries, the future of AI development and the rights of millions of content creators.
Why Is Generative AI Creating a Copyright Crisis?
Generative AI systems learn patterns from enormous collections of information. These datasets can contain books, photographs, illustrations, music, news articles, websites, films, computer code and other creative material.
During training, AI developers may make digital copies of source material, process it and use it to adjust the parameters of a machine learning model. The model can then generate new text, images, audio or other outputs in response to user instructions.
The controversy is that some of the material used to train these systems may be protected by copyright. Rights holders argue that their work has been copied and commercially exploited without authorization or payment.
Books and written content
Writers argue that their books, articles and other publications have been used to train chatbots and writing assistants without appropriate licenses.
Images and illustrations
Photographers and illustrators question whether image generators can learn from their portfolios without permission and produce competing images.
Songs and recordings
Music publishers and recording companies challenge the use of lyrics, compositions and sound recordings to train AI music generators.
Code and programming
Developers raise questions about code licensing, attribution, open-source conditions and AI-generated code that resembles existing software.
These disputes involve several separate legal questions. Using a copyrighted work during training is not necessarily the same legal issue as generating an output that reproduces protected expression. Copyright ownership of the output is another distinct question.
Can AI Companies Train Models on Copyrighted Works?
This is one of the most important unresolved questions in generative AI law. There is no universal answer because copyright exceptions differ between countries and even between types of use.
The argument for fair use
In the United States, AI developers often rely on the fair use doctrine. Section 107 of the US Copyright Act allows certain uses of copyrighted material without permission, depending on the circumstances.
Courts generally consider four factors:
- Purpose and character of the use: Whether the new use is transformative, commercial or serves a different purpose from the original.
- Nature of the copyrighted work: Whether the source material is factual or highly creative.
- Amount used: How much of the original work was copied in relation to the purpose of the use.
- Market effect: Whether the use harms existing or potential markets for the original work.
AI companies argue that training transforms material into statistical patterns and does not simply redistribute the original books, images or recordings. They also contend that requiring permission for every training item could make large-scale AI development impractical.
The argument against unlicensed training
Copyright owners respond that copying millions of protected works to build commercial AI products can interfere with their exclusive rights. They argue that AI systems may compete directly with the creators whose material helped develop them.
Rights holders also raise concerns about unauthorized downloading, the use of pirated books and images, the removal of copyright information and outputs that reproduce recognizable portions of protected works.
Why the source of training data matters
One of the important distinctions emerging from US litigation is the difference between acquiring a work lawfully and obtaining it from an unauthorized source.
A court may conclude that a particular use of lawfully acquired material is transformative while reaching a different conclusion about maintaining a library of pirated copies. The legal analysis depends on the specific conduct, not simply on the fact that an AI model was trained.
The Major Generative AI Copyright Lawsuits
Several high-profile disputes are shaping the legal debate. Some have produced significant rulings or settlements, while others remain unresolved. The following examples illustrate the different legal theories being tested.
1. Bartz v. Anthropic: A major settlement over books
Authors brought a class action against Anthropic, alleging that copyrighted books had been used without authorization in the development of its AI systems.
In June 2025, a US federal district court ruled that training Claude on books Anthropic had lawfully purchased and digitized was fair use on the record before it. However, the court distinguished that conduct from downloading pirated books and maintaining a central library of unauthorized copies.
In July 2026, the court granted final approval to a settlement worth approximately $1.5 billion. The settlement covers a large class of works and represents one of the most significant financial developments in AI copyright litigation.
The settlement should not be interpreted as a universal ruling that all AI training is unlawful. It resolved claims in that specific case rather than establishing a rule for every AI developer or dataset.
2. Kadrey v. Meta: Fair use and the importance of market evidence
A group of authors sued Meta over the use of their books to train its Llama AI models. The plaintiffs argued that their copyrighted works had been copied without permission.
In June 2025, the district court ruled in Meta’s favor on the fair use issue presented in the case. The judge considered the training transformative and found that the authors had not developed sufficient evidence of market harm on the record before the court.
The decision highlighted a major challenge for copyright owners: showing that an AI product competes with or damages the market for their original work may require detailed economic evidence.
The ruling did not establish that every use of pirated material is lawful or that all AI training automatically qualifies as fair use.
3. The New York Times v. OpenAI and Microsoft
The New York Times filed a major copyright lawsuit against OpenAI and Microsoft in December 2023. The newspaper alleges that its articles were used without authorization to train AI systems and that some outputs can reproduce or closely summarize its journalism.
The dispute raises questions about whether AI chatbots can become substitutes for news publications, how copyrighted articles are represented in training datasets and whether outputs can reproduce protected expression.
As of November 2026, the litigation remains a major unresolved test of the relationship between generative AI and the commercial market for journalism. The case could influence how publishers negotiate licensing agreements with technology companies.
4. Andersen v. Stability AI: Artists challenge image generators
Visual artists have challenged AI image-generation companies, including Stability AI, over the alleged use of copyrighted artwork in training datasets.
The artists argue that image generators were developed using their work without permission and can produce images that imitate recognizable artistic styles or reproduce protected elements.
These disputes raise difficult questions about the distinction between copying an actual protected work and learning general artistic characteristics. Copyright generally protects original expression, not an artistic style in the abstract.
As of November 2026, the central questions concerning the use of photographs and illustrations to train image-generating models remain contested in major US and UK litigation.
5. Getty Images v. Stability AI
Getty Images has pursued legal claims against Stability AI concerning the alleged use of its image library in the development of Stable Diffusion.
The UK proceedings have involved questions about copyright, database rights and trade marks. The original UK trial did not produce a broad ruling that resolved the legality of all image-model training. The training claim was not decided on its merits in that proceeding.
An appeal concerning aspects of the case was heard by the Court of Appeal of England and Wales in November 2026, with judgment pending at the time of this article.
6. GEMA v. Suno: Music rights and AI-generated songs
Music copyright disputes are particularly complicated because a song can involve multiple protected elements, including musical compositions, lyrics and sound recordings.
In July 2026, the Munich Regional Court ruled in favor of German music collecting society GEMA in a case against AI music generator Suno. The court found that protected musical works had been reproduced in connection with the system and addressed the use of those works in generated outputs.
The decision is an important European development, although its implications must be understood in the context of German and EU copyright law and any subsequent appeal.
7. ANI v. OpenAI: An important Indian dispute
Indian news agency ANI brought a copyright lawsuit against OpenAI concerning the alleged use of its articles in AI training and related issues.
In July 2026, the Delhi High Court declined to grant ANI an interim injunction. The court expressed a preliminary view that the training use could fall within the fair dealing framework on the record before it, while also considering the question of whether AI outputs reproduced the agency’s protected material.
This was an interim-stage development, not a final resolution of every copyright claim. The case remains relevant to the international debate about how national copyright exceptions apply to generative AI.
Key AI Copyright Cases: Quick Comparison
| Case | Main issue | Position as of November 2026 |
|---|---|---|
| Bartz v. Anthropic | Books used to train AI models | Important fair use ruling on lawfully acquired books; separate settlement approved for approximately $1.5 billion. |
| Kadrey v. Meta | Books used to train Llama | Meta prevailed on fair use in the district court ruling, with significant emphasis on the evidence of market harm. |
| New York Times v. OpenAI and Microsoft | News articles, training and outputs | Major litigation continues; market substitution and reproduction issues remain important. |
| Andersen v. Stability AI | Artists’ works used for image generation | Central training and output questions remain contested. |
| Getty Images v. Stability AI | Images, databases and trade marks | UK appeal heard in November 2026; judgment pending. |
| GEMA v. Suno | Music works and AI-generated songs | German court ruled for GEMA in July 2026; further legal developments may follow. |
| ANI v. OpenAI | News content and AI training in India | Interim injunction refused in July 2026; litigation continues. |
Case status is a snapshot, not a prediction of final outcomes. Settlements, interim decisions and trial-level judgments should not be confused with final appellate precedent.
Generative AI and Copyright Laws Around the World
Different legal systems are approaching AI training and copyrighted material in different ways. Some rely on existing copyright exceptions, while others are considering specific transparency, licensing or compensation requirements.
United States: Fair use is central to the debate
The US Copyright Act does not contain a specific blanket exception for generative AI training. Developers often rely on fair use, but the outcome depends on the facts and the four statutory factors.
Recent litigation shows that courts may distinguish between lawful acquisition, unauthorized copying, training purposes and outputs that compete with original works.
The US Copyright Office has also examined AI training and human authorship. Its reports and guidance are important resources, although they do not replace decisions by courts in individual disputes.
European Union: Copyright exceptions and transparency obligations
The EU has a legal framework that includes copyright rules on text and data mining (TDM), alongside the AI Act and other digital legislation.
The Copyright in the Digital Single Market Directive provides text and data mining exceptions under specific conditions. For general-purpose commercial mining, rights holders can reserve their rights in an appropriate manner. The rules therefore do not provide unlimited permission to use every copyrighted work for AI training.
The EU AI Act introduces transparency obligations for providers of general-purpose AI models, including a policy to comply with EU copyright law and a sufficiently detailed summary of training content. These obligations are not equivalent to a universal licensing requirement for every work.
United Kingdom: Text and data mining remains a major policy issue
The UK has debated whether copyright exceptions should be expanded to support AI development and how creators should be able to control the use of their work.
Existing UK copyright rules include a text and data mining exception for certain non-commercial research purposes. Commercial AI training cannot simply be assumed to fall within that exception.
Policy discussions continue to examine licensing, transparency, creator control and the economic impact of AI-generated content.
China: Copyright law and AI-generated content
China’s copyright framework applies to protected creative works and has been tested in disputes involving AI-generated images and content.
Chinese courts have considered whether certain AI-generated outputs can qualify for copyright protection when a person makes sufficiently original intellectual contributions. The analysis depends on the human involvement and the specific work.
China also regulates AI services through rules that address security, content management and other obligations. These regulations should be distinguished from the separate question of copyright ownership and infringement.
India: Existing copyright exceptions under scrutiny
India’s Copyright Act, 1957, contains fair dealing provisions and specific exceptions. The ANI dispute illustrates how courts may be asked to interpret those provisions in the context of machine learning and generative AI.
There is no simple universal permission allowing every AI developer to use all copyrighted works for commercial training. The legal position depends on the applicable provisions and facts.
Latin America: Existing copyright laws face new challenges
Countries across Latin America are assessing generative AI through existing copyright legislation, data protection rules and emerging policy discussions.
Brazil, Mexico, Argentina, Chile, Colombia and Peru have different legal frameworks. None should be treated as having an identical AI training exception or licensing regime.
For creators and businesses in the region, important questions include whether source material was lawfully obtained, whether an output reproduces protected expression, and whether existing national exceptions apply.
Who Owns AI-Generated Content?
Another major legal question concerns ownership. If a person types a prompt into an AI system and receives a photograph, illustration, article or song, does that person automatically own the copyright?
The answer depends on the jurisdiction and the level of human creative contribution.
Human authorship in the United States
The US Copyright Office has maintained that copyright protection requires human authorship. A work created entirely by an AI system without sufficient human creative contribution generally cannot be registered as a human-authored copyrighted work.
However, a human-created work does not necessarily lose protection simply because AI tools were used during the creative process. Human-authored elements, original selection and arrangement, and sufficiently creative modifications may qualify for protection.
Why prompts alone may not be enough
A simple text instruction may not demonstrate enough human control over the expressive elements of the final output. The more a person determines, selects, modifies and arranges the protected expression, the stronger the potential argument for human authorship.
Copyright analysis is fact-specific. The length or complexity of a prompt alone does not automatically establish ownership.
What about AI-generated music and images?
Similar questions apply to music and visual art. If a person uses AI to generate a song and then makes substantial original compositional changes, some human-created elements may qualify for protection. A fully automated image may receive different treatment.
Human-created work
A person writes an article or paints an image without using generative AI. Traditional copyright rules generally apply.
AI-assisted work
A person uses AI for brainstorming, editing or technical assistance but contributes original human expression. Protectable human elements may remain.
AI-generated output
A system produces the expressive content with little or no meaningful human authorship. Copyright protection may be unavailable in jurisdictions requiring human authorship.
It is also important to distinguish copyright ownership from contractual rights. An AI platform’s terms of service may grant users certain rights or permissions, but a contract cannot automatically create copyright protection where the law does not recognize an author.
How Generative AI Affects Artists, Writers and Musicians
Generative AI creates new opportunities for creative professionals, but it also changes the economics of content production.
Competition from AI-generated content
Businesses can use AI systems to produce marketing copy, illustrations, background music and other material at a lower cost. This can increase competition for freelance writers, photographers, designers and musicians.
Creators argue that AI companies should not be able to use their work to build competing commercial products without permission or compensation.
Style imitation and creative identity
Artists have raised concerns about AI systems that can generate images resembling recognizable artistic styles. Copyright generally protects specific original expression rather than an abstract style, but other legal issues may arise depending on the circumstances.
For example, trade mark law, false endorsement, publicity rights or unfair competition rules may be relevant when an AI-generated work misleadingly suggests that a particular artist created or endorsed it.
Music, voices and digital replicas
AI-generated songs can imitate vocal characteristics or create performances that sound similar to known artists. This raises issues beyond traditional musical copyright.
Depending on the jurisdiction, unauthorized digital replicas may involve rights of publicity, privacy, performer protections, contract law or specific legislation addressing synthetic replicas.
New opportunities for independent creators
AI tools can also help smaller creative businesses reduce production costs, develop concepts, translate material and experiment with new formats.
The long-term effect will depend partly on whether creators can negotiate licensing agreements, receive attribution where appropriate and retain meaningful control over their work.
The Rise of AI Copyright Licensing Agreements
As litigation increases, licensing is emerging as a potential alternative to prolonged court disputes.
AI developers may negotiate agreements with publishers, media organizations, image libraries, music rights holders and other content owners. These arrangements can provide access to material under defined commercial terms.
What can an AI licensing agreement include?
- Authorized content: A clear description of the books, articles, images, recordings or datasets covered.
- Permitted uses: Whether the material can be used for training, fine-tuning, retrieval or other purposes.
- Payment structure: A fixed fee, subscription, usage-based payment, revenue share or another compensation model.
- Attribution: Whether the content owner receives credit or visibility in certain outputs.
- Data retention: Rules for storing, deleting or updating licensed material.
- Output restrictions: Conditions addressing reproduction of protected works or other prohibited uses.
- Audit rights: Mechanisms for checking whether the license is being followed.
Why publishers and technology companies are negotiating
For publishers, licensing can create a new source of revenue and provide greater control over the use of their material. For AI developers, it can reduce uncertainty around data acquisition and create more predictable commercial relationships.
However, licensing does not resolve every legal question. A license from one publisher may not cover all authors, photographers, contributors or other rights holders whose material appears in a dataset.
Some agreements may also raise questions about whether creators have authority to license the relevant rights and whether the compensation reaches the original rights holders.
What Businesses Should Know About AI and Copyright
Companies using generative AI should develop clear policies before incorporating AI-generated material into websites, advertising campaigns, software products, books or commercial projects.
1. Check the terms of the AI platform
Review the platform’s terms of service, commercial-use permissions, indemnification provisions and restrictions. Do not assume that every AI tool provides the same rights.
2. Keep records of human contributions
Document the creative process, including original drafts, edits, design decisions, revisions and human contributions. This can help establish which elements were created by people.
3. Avoid requesting direct copies of protected works
Do not instruct AI systems to reproduce entire copyrighted articles, illustrations, songs, photographs or other protected material without authorization.
4. Verify generated content
Review outputs for substantial similarity to existing works, unauthorized logos, copied text, recognizable characters and other potentially protected elements.
5. Review training and source data when developing AI products
Organizations building their own models should maintain records of dataset sources, licensing terms, permissions and applicable copyright exceptions.
6. Establish a clear AI-use policy
Employees and contractors should understand which tools they can use, what information they can upload and who is responsible for checking the legal status of commercial outputs.
7. Review contracts with customers and suppliers
Contracts should clarify responsibility for AI-generated content, ownership of human-created contributions, warranties, indemnities and procedures for responding to copyright claims.
AI Copyright Risk Checklist for Businesses
Review these questions before publishing or selling AI-generated content.
The Next Decade: Five Legal Battles That Could Define AI Copyright
1. Will AI training require permission?
The biggest question is whether commercial AI developers will be able to rely on existing copyright exceptions or whether lawmakers will require more licensing and compensation. Different jurisdictions may reach different conclusions.
2. How will courts measure market harm?
Courts will need to assess whether AI products replace demand for books, journalism, music, illustrations or other creative works. This could require new economic evidence and more detailed analysis of how AI-generated content competes with human-created material.
3. Will creators receive compensation?
Licensing arrangements, collective rights management and possible statutory mechanisms may create new ways for creators to receive payment when their work is used in AI development.
4. How will copyright apply to AI-generated outputs?
Courts and copyright offices will continue to address the level of human involvement necessary for protection. Businesses will also need to understand the difference between contractual permissions and actual copyright ownership.
5. Who is responsible when AI reproduces protected material?
Future cases may clarify the responsibility of AI developers, platform operators, distributors and users. Liability could depend on how a model was trained, how it was designed, what safeguards were implemented and how the output was used.
6. Will international rules become more consistent?
AI systems operate across borders, but copyright remains largely territorial. A model may be trained in one country, hosted in another and used to generate content for customers worldwide. International cooperation and clearer rules may become increasingly important.
The next decade will likely produce a more defined legal framework, but it is unlikely that every country will adopt the same approach. Businesses operating internationally will need to monitor several legal systems at once.
Frequently Asked Questions About Generative AI and Copyright
Can AI companies legally use copyrighted material to train their models?
It depends on the country, the source of the material, the purpose of the use and the applicable copyright exceptions. In the United States, companies often argue that training qualifies as fair use. European rules include specific text and data mining exceptions with conditions. There is no universal permission covering all copyrighted material.
Is AI-generated content protected by copyright?
It depends on the jurisdiction and the amount of human creative contribution. In the United States, purely AI-generated material without sufficient human authorship generally cannot receive copyright protection. Human-created elements and sufficiently creative modifications may qualify.
Can I copyright an image created with an AI tool?
Possibly, depending on the applicable law and your creative contribution. A person who meaningfully modifies, selects or arranges AI-generated material may have copyright in the human-created elements, even if the automatically generated portions are not independently protected.
Can artists sue AI companies for using their work?
Artists and other rights holders can bring legal claims when they believe their protected works have been copied or used unlawfully. Whether a claim succeeds depends on the facts, evidence, applicable copyright law and any relevant exceptions or defenses.
What happened in the Anthropic copyright lawsuit?
In Bartz v. Anthropic, a US court found that training on lawfully purchased and digitized books was fair use on the record before it, while distinguishing the maintenance of pirated copies. A settlement worth approximately $1.5 billion received final court approval in July 2026.
Does fair use mean AI companies can use anything they find online?
No. Fair use is a fact-specific legal defense, not a general license to copy any online material. The source of the content, the purpose of copying, the amount used and the effect on the market can all matter.
Can an AI-generated song infringe copyright?
Yes. An AI-generated song may raise copyright issues if it reproduces protected lyrics, melodies, recordings or other protected expression. Voice imitation and digital replicas can also raise separate legal issues.
Who owns content generated by ChatGPT or another AI platform?
The answer depends on the platform’s terms, applicable law and the human contribution involved. A platform may grant users certain contractual rights, but those terms do not automatically establish copyright protection for material that is not legally copyrightable.
Will AI companies have to pay authors and artists?
Some companies are negotiating licensing agreements with publishers, music companies and other rights holders. Courts and lawmakers may also shape future compensation systems. There is no single global payment rule that applies to every AI training dataset.
Can businesses use AI-generated content for commercial purposes?
Many AI platforms permit certain commercial uses under their terms, but permission from a platform does not guarantee that an output is free from third-party copyright claims. Businesses should review the terms, verify outputs and assess applicable laws before publication.
Conclusion: The Future of AI Depends on the Future of Copyright
Generative AI and copyright are on a collision course that will shape the technology industry and creative economy for years to come. The legal disputes involving Anthropic, Meta, OpenAI, Stability AI, Suno and other companies are testing fundamental questions about how creative works can be used in machine learning.
Some court decisions have supported particular AI training uses, while others have emphasized the risks of unauthorized source material, market competition and reproduction of protected works. Major disputes remain unresolved, and settlements do not establish universal legal rules.
For creators, the challenge is protecting their work while adapting to new production technologies. For AI developers, the challenge is building systems with lawful data practices and sustainable relationships with content owners. For businesses, the safest approach is to understand platform terms, document human contributions and review generated material before using it commercially.
The next decade will determine whether generative AI develops primarily through broad copyright exceptions, negotiated licensing markets, new legislation or a combination of all three.
Sources and Further Reading
These resources provide additional information about copyright law, AI training and the major legal disputes discussed in this article.
- US Copyright Office – Copyright and Artificial Intelligence
- US Copyright Office – Fair Use
- US Congress – Copyright legislation and policy resources
- Congressional Research Service – Generative Artificial Intelligence and Copyright Law
- European Union – Copyright in the Digital Single Market Directive
- European Union – Artificial Intelligence Act
- World Intellectual Property Organization – Copyright
- CourtListener – US court opinions and legal filings
Disclaimer: This article is for general informational purposes and does not constitute legal advice. Copyright rules vary by jurisdiction and may change as new court decisions and legislation emerge. Consult qualified legal counsel before using copyrighted material to train AI systems or commercializing AI-generated content.