What Have Courts Ruled with Respect to AI and Copyright?

The emergence of generative artificial intelligence (AI) has revolutionised content creation. These AI models learn by ingesting massive amounts of data from the internet, including copyrighted works. This practice has raised important legal questions: Does using copyrighted material to train AI infringe copyright laws? Or can such use be justified under existing exceptions like fair use or fair dealing? Courts worldwide have begun addressing these complex issues, but the law is still evolving.

In 2025, three major cases from the United States—Thomson Reuters v. Ross Intelligence, Bartz v. Anthropic, and Kadrey v. Meta—have brought some clarity. This article examines these judgments, explores their implications, and looks at the current legal landscape in India regarding AI and copyright.

Understanding AI Training and Copyright Issues

Generative AI models require large datasets to ‘learn’ how to generate text, images, or other creative outputs. These datasets often include:

• Public domain content, which is free to use.
• Copyrighted material, which is protected and requires permission for use unless an exception applies.

The central legal question is whether copying copyrighted works for AI training constitutes copyright infringement.

Most jurisdictions recognise some exceptions to copyright infringement for purposes such as research, criticism, or education. In the United States, this is known as the fair use doctrine, which allows limited use of copyrighted material without permission under certain conditions. In the European Union and the United Kingdom, specific text and data mining (TDM) exceptions exist.

However, applying these exceptions to AI training is not straightforward, especially when the data includes pirated or illegally obtained material. Moreover, the question of whether AI-generated content itself can be copyrighted remains unsettled.

Key U.S. Court Judgments on AI and Copyright

The three key U.S. cases in 2025 provide important judicial perspectives on AI training with copyrighted works.

Thomson Reuters v. Ross Intelligence

Ross Intelligence, an AI company, was accused of unlawfully using Thomson Reuters’ copyrighted legal publications to train its AI models.

The court held that the training was transformative, comparing it to a student learning from textbooks. This transformative nature qualified as fair use.

This ruling affirmed that AI models have the right to learn from copyrighted works, provided that the new work adds value and does not simply copy the original.

Bartz v. Anthropic

Anthropic faced allegations of training its AI with copyrighted books and articles, some of which were pirated copies.

Judge William Alsup ruled that training AI is akin to human learning and therefore transformative. Hence, it can fall under fair use.

However, the use of pirated content raised a separate legal issue. Anthropic was ordered to face trial to determine the legality of using illegally sourced material.

This case clarified that while transformative use may be protected, unlawful acquisition of data remains actionable.

Kadrey v. Meta

In this case, the plaintiffs argued that Meta’s AI-generated content diluted the market for their original works.

Judge Vince Chhabria ruled in favour of Meta, finding insufficient evidence of market harm or substitution.

Meta’s use of copyrighted materials for AI training was deemed fair use, even though the company monetised AI products.

However, the court emphasised that tech firms benefiting from AI should consider mechanisms to share revenue with copyright holders.

Public Domain vs. Copyrighted Works: Legal Distinctions

AI training on public domain material carries no legal risk. Using copyrighted works, however, requires careful legal justification, especially if the AI output closely resembles the original.

Global Regulatory Landscape and Ambiguities

No global harmonisation exists on how copyright applies to AI training. Jurisdictions vary:

• United States: Flexible fair use doctrine is often invoked, but its application to AI is still developing.
• European Union and United Kingdom: Specific text and data mining exceptions allow non-commercial research but limit commercial use.
• India: Follows a “fair dealing” exception, which is narrower than fair use.

Key unresolved questions worldwide include:

• Who owns AI-generated content? Can it attract copyright protection?
• Does copying large datasets for AI training infringe copyright?
• Should creators be compensated when AI uses their works?

The Indian Legal Framework on AI and Copyright

India’s copyright law, governed by the Copyright Act, 1957, provides:

• Section 14: Exclusive rights to reproduce, adapt, translate, or communicate a work.
• Section 52: Fair dealing exceptions for purposes like research, criticism, or reporting.

Currently, India has no AI-specific copyright provisions. Courts may interpret fair dealing exceptions to cover AI training, but this is untested.

India’s fair dealing is more restrictive than U.S. fair use, focusing on specific purposes and lacking a flexible market effect analysis.

Additionally, Indian law requires human creativity for authorship. Whether AI-generated works qualify for copyright is an open question.

The ANI v. OpenAI Case: A Crucial Indian Lawsuit

Recently, Aadhaar News & Information (ANI) sued OpenAI alleging unauthorised use of ANI’s copyrighted news footage in AI training.

This case is significant for India as it will test:

• Whether India’s existing copyright law can regulate AI training.
• How courts will balance fair dealing exceptions with copyright owners’ rights.
• Whether commercial AI companies require licences for copyrighted training data.

The case is closely watched for its potential to shape India’s AI and copyright policy.

Conclusion

Courts in the United States have begun to clarify that training AI models on copyrighted works can qualify as fair use if the use is transformative and does not cause market harm. However, unauthorised use of pirated material remains unlawful, and the question of fair compensation to rights holders is unresolved.

India’s copyright law currently lacks AI-specific provisions, and the outcome of cases like ANI v. OpenAI will be pivotal. For now, AI developers and copyright owners alike must navigate a complex and evolving legal landscape, balancing innovation with respect for intellectual property rights.