Context

• Artificial Intelligence (AI) is transforming how people work, create, and communicate. AI models like ChatGPT and Gemini are trained on enormous datasets containing books, articles, images, and sounds collected from the internet. While this data is crucial for making AI systems capable of generating text, images, or music, it has also sparked serious legal and ethical debates.
• Many authors, artists, and companies argue that training AI on copyrighted materials without permission amounts to theft. They fear that AI could produce content similar to their original work, reducing demand for authentic creations and harming their livelihoods.
• At the same time, tech companies claim that their use of data is “transformative.” They argue that AI learns patterns from material and creates something new, which should be allowed under the doctrine of “fair use.” Fair use is a legal principle that permits limited use of copyrighted material without permission in specific cases like criticism, commentary, teaching, and research.
• Recently, two important court cases in the United States ruled in favour of AI companies, marking the first legal decisions on whether training AI models on copyrighted works is permissible. These judgments set important precedents but also left many questions unanswered.

Understanding Fair Use and Transformative Use:

Fair use is central to the legal debate over AI and copyright. Under this principle, courts decide whether using copyrighted material without authorization is acceptable by evaluating factors such as:

• The purpose and character of the use
• The nature of the copyrighted work
• The amount and significance of the portion used
• The effect of the use on the market for the original work

When an AI model uses copyrighted material to learn patterns and generate entirely new content, tech companies argue this is transformative use. Transformative use means the work has been changed in purpose, meaning, or character so much that it becomes a new form of expression rather than a copy.

These arguments were central in both recent cases. While the courts accepted that AI can be transformative, they also recognized that using pirated data creates further legal complications.

Legal Precedent: AI Training and Copyright Use

Court Ruling on Use of Books in AI Training

A group of writers filed a class-action lawsuit against an AI company, alleging that it had trained its large language models using pirated books without the authors’ consent. The books were reportedly sourced from Books3, an online shadow library containing millions of copyrighted texts.

Key Allegations by the Writers:

• The company copied pirated versions of their books to train its models.
• This reduced their income and undermined the market for original written content.
• The AI-generated outputs could replicate the type of work that authors usually get paid to create.

Defense by the AI Company:

• The company admitted to using Books3 but also said it legally purchased and scanned millions of printed books to build a broader dataset.
• It claimed the use was transformative, similar to a person reading books to become a better writer.

Court’s Decision:

• The judge ruled that the training of AI on copyrighted material could be considered fair use because the outputs were new and different.
• He emphasized that the AI models “turn a hard corner and create something different,” rather than replicate the original works.
• However, the court also acknowledged that storing and copying pirated materials infringes copyright.
• A separate trial has been ordered to determine how much the company owes in damages for that infringement.

Court Ruling on Training Methods and Market Impact:

Another group of writers filed a class-action suit against a different AI company, claiming that its language models were trained on copyrighted content without permission, sourced from shadow libraries like Books3, Anna’s Archive, and Libgen.

Key Concerns of the Writers:

• The authors argued that the AI’s ability to generate content based on their books reduced the demand for original works.
• They sought financial damages and compensation for market dilution.

Defense by the AI Company:

• The company said it took steps to ensure its models did not reproduce large portions of any specific copyrighted text.
• It claimed that internal tests showed the models could not generate more than 50 consecutive words from any copyrighted book.
• The company also argued that the outputs did not harm the market for original books.

Court’s Decision:

• The judge noted that the plaintiffs had not provided evidence of market dilution.
• He stated that unless the AI-generated outputs replaced demand for original books, the market effect was not proven.
• While accepting the AI’s transformative nature, the court also mentioned that tech firms benefiting from AI should consider ways to compensate creators.
• The company must still face further legal proceedings regarding the use of pirated content.

Ongoing Legal Disputes and Wider Implications

While the recent rulings favoured AI companies on the question of fair use, they did not provide full legal clearance. Both companies are still being held accountable for the use of pirated material.

Moreover, many other lawsuits are active or emerging:

• Twelve consolidated lawsuits from authors, publishers, and news outlets against OpenAI and Microsoft.
• Visual artists suing image-generation tools for using their art without consent.
• Getty Images suing Stability AI for using over 12 million of its copyrighted photographs.
• In India, a major news agency and several digital publishers have sued OpenAI for using Indian content in training without permission.

These cases show that legal challenges to AI’s training practices are just beginning. Concerns around fair use, creator compensation, and piracy will likely dominate the policy agenda in the coming years.

Conclusion:

The recent US court rulings highlight the complexity of balancing innovation with the rights of creators. While AI companies argue that training models on vast datasets is essential for progress and serves the public good, authors and artists see this as exploitation of their work without acknowledgment or compensation.

The principle of fair use, especially transformative use, has become the cornerstone of these legal battles. However, judges have recognized that while AI models may create something new, using pirated content still raises serious legal and ethical concerns.

Moving forward, clear regulations and frameworks are needed to define:

• What counts as fair use in AI training
• How creators can be fairly compensated
• How to prevent misuse of copyrighted content
• What safeguards should be in place to protect markets for original works

These issues will not be resolved quickly. As AI systems continue to evolve, courts, policymakers, and industry stakeholders will need to work together to create fair rules that encourage innovation while respecting the rights of authors, artists, and other creators. The outcomes of these cases will shape how AI and copyright law develop in the coming years, both in the US and around the world.