Fair Use on Trial: What Meta Decision Means for Copyright and AI Governance

A United States court in San Francisco recently ruled that Meta’s use of books to train its artificial Llama AI models constituted fair use under copyright law. The lawsuit, filed by a group of authors including award-winning author and journalist Ta-Nehisi Coates, challenged Meta’s alleged use of materials sourced from Library Genesis (LibGen), an online repository of books, academic articles, and comics.

While the court sided with Meta, the judgment was mainly based on the plaintiffs’ failure to sufficiently argue their case, rather than establishing a definitive legal precedent that copyrighted works can be freely used to train AI. Despite this, the case highlights the broader legal and ethical questions surrounding the use of copyrighted content in the development of artificial intelligence, which is built upon data as its primary building block.

One school of thought argues that training AI on publicly available or copyrighted content falls under the doctrine of fair use. In the United States, this doctrine allows limited use of copyrighted material without the rights holder’s permission under certain circumstances.

In Tanzania, a similar principle is reflected in Section 12 of the Copyright and Neighbouring Rights Act, 1999, which addresses “Free Use” of copyrighted works. The provision outlines specific situations where copyrighted material may be used without obtaining permission from the rights holder. These include use for research or private study, criticism, review, or reporting of current events, as well as educational purposes such as classroom teaching and judicial proceedings.

Proponents of fair use in the Artificial Intelligence realm argue that AI does not reproduce or republish content in its original form; instead, it learns patterns to generate novel outputs, much like people learn by reading and observing. Under this perspective, restricting access to data would stifle innovation, entrench existing tech monopolies and prevent open research.

In contrast, another school of thought argues that such data use constitutes unauthorised exploitation of copyrighted works, particularly when those materials are used on a large scale without compensation or attribution. Critics say that allowing AI developers to scrape books, articles, music, and artwork effectively devalues creative labour, undermines the rights of content creators, and creates an uneven playing field. Some go further to suggest that AI-generated outputs trained on copyrighted data are “derivative works,” which should require licensing.

This argument was aptly captured by the presiding Judge Vince Chhabria in the Meta Llama case, who stated: “People can prompt generative AI models to produce these outputs using a tiny fraction of the time and creativity that would otherwise be required…This could dramatically undermine the incentive for human beings to create things the old-fashioned way.”

While both sides of the debate present fair arguments, emerging AI governance frameworks have taken different approaches to this dilemma. The European Union’s AI Act does not contain specific provisions on intellectual property rights. Instead, it defers to existing legislation, including the 2019 EU Copyright Directive, which provides a legal exception for text and data mining under certain conditions.

The AI Act focuses primarily on transparency, accountability, and risk management in the development and deployment of AI systems. It requires providers of high-risk and general-purpose AI models to disclose whether copyrighted or personal data was used in training, and to implement safeguards to mitigate misuse or harm. However, the Act does not establish new copyright exceptions, such as fair use or fair dealing, instead relying on the existing copyright framework across EU member states.

The United States currently lacks a comprehensive federal law governing AI. However, courts’ rulings on cases involving the use of copyrighted materials for AI training often rely on the fair use doctrine.

Kenya’s recently published Artificial Intelligence Strategy (2025–2030) acknowledges the need to revise intellectual property laws to align with the demands of AI. However, it does not clearly outline how intellectual property rights should interact with the development of AI. While the strategy does not provide detailed guidance on the use of copyrighted materials for AI training, it calls for a balanced approach that encourages innovation while respecting existing legal frameworks, including those governing copyright and data protection.

Tanzania’s Draft Strategy for the Responsible Use of Artificial Intelligence also highlights concerns related to the use of AI, such as the protection of intellectual property rights. It has therefore identified legal and regulatory reforms as one of the enablers of achieving its artificial intelligence goals. The Tanzanian government, therefore, aims to conduct a comprehensive review of existing regional AI policies, identify gaps and develop new AI policies and guidelines.

From current trends, it appears that most jurisdictions are leaning toward adopting fair use or similar principles to support the development of AI models and solutions in the interest of national progress. However, as countries strive to establish their presence in the global AI landscape by harnessing its transformative potential, it is crucial to develop clear guidelines for the use of copyrighted works. These guidelines should strike a balance between enabling innovation and respecting the IP rights of authors and creators.