Legal Battles and Licensing Deals Shape Copyright’s Role in AI Development

On September 23, 1962, "The Jetsons" premiered, envisioning a future where technology dramatically reshapes daily life. While the series projected a 2060 utopia, many aspects of its vision resonate today, particularly with the rise of artificial intelligence (AI) and generative AI technologies promising unprecedented efficiency and creativity.

AI’s rapid integration into diverse fields—from companionship and advertising to legal drafting—has sparked both enthusiasm and resistance. Despite some advocating for digital minimalism, AI’s momentum continues, raising critical questions about the role of intellectual property (IP) in this new paradigm.

In the billion-dollar AI industry, information is paramount, and intellectual property rights remain central. AI systems depend heavily on human-created content—including copyrighted and non-copyrighted materials—to train, learn, and generate outputs. Given that copyright law traditionally protects human authorship by granting exclusive rights for a limited time, the use of copyrighted works in AI training has become a contentious issue.

The legal community is divided on whether and how copyrighted materials should be utilized in AI development. Without clear, harmonized legislation, some rights holders have pursued litigation to establish precedent. They argue that training AI models on copyrighted works without permission constitutes infringement. Conversely, AI developers claim such use qualifies as "fair use," emphasizing that restricting access to publicly available content would hinder creativity, economic growth, and cultural exchange.

Recent case law tends to favor copyright holders. In Thomson Reuters Enterprise Center GmbH v. Ross Intelligence Inc., the Delaware District Court ruled that Ross Intelligence’s use of over 2,000 Westlaw headnotes to train its AI legal research tool infringed Thomson Reuters’ copyrights. The court found the headnotes sufficiently original to warrant protection and determined that Ross’s commercial, non-transformative use negatively impacted Thomson Reuters’ market, outweighing fair use defenses.

In Kadrey v. Meta Platforms, Inc., thirteen authors alleged that Meta trained its large language models using their copyrighted books obtained from shadow libraries. The Northern District of California granted summary judgment to Meta but clarified that the ruling was narrow and did not broadly legalize Meta’s use of copyrighted materials for AI training. The court found Meta’s use transformative and unlikely to harm the authors’ market, though it acknowledged the theoretical potential for AI-generated works to compete with originals.

Some AI companies opt to settle rather than litigate. Anthropic, a leading AI developer, faced lawsuits claiming it used pirated books from sites like Library Genesis to train its flagship model, Claude. In June, a court issued a nuanced summary judgment recognizing both transformative use and infringement, rejecting fair use as a defense. Anthropic subsequently agreed to a $1.5 billion settlement, preliminarily approved by Judge William Alsup in September.

Beyond litigation, some rights holders pursue licensing agreements with AI firms. Disney’s recent three-year deal with OpenAI exemplifies this approach, providing licensed access to copyrighted works for AI training. Such partnerships offer licensors new revenue streams and licensees authorized, high-quality training data, potentially enhancing AI output quality. However, negotiations must carefully address data ownership, licensing scope, derivative works, and valuation of AI-related rights.

Copyright questions also arise regarding AI-generated content. The U.S. Copyright Office permits copyright registration for works incorporating AI under limited conditions, requiring significant human creative input and control. For example, registrations have been granted for a comic book with AI-generated images, a visual work created with Invoke AI, and a music opera featuring AI-generated voices. In each case, the office excluded purely AI-generated elements from protection, focusing on human selection, arrangement, and coordination.

Importantly, the Copyright Office maintains that AI itself cannot be an author under current law. This principle was upheld in Thaler v. Perlmutter, where the D.C. Circuit affirmed the denial of copyright for a work created solely by an AI system named "Creative Machine." The court emphasized that human authorship is a prerequisite for copyright protection.

Currently, the evolving legal framework suggests that explicit permission from rights holders may soon be necessary for using copyrighted materials in AI training. Meanwhile, works created with AI can receive copyright protection if human authorship is sufficiently demonstrated, albeit with nuanced limitations.

Beyond legal and commercial considerations, broader societal questions persist: How much should AI influence our lives? What safeguards are necessary? What is the impact on workforce displacement, especially in creative industries historically supported by IP law? These issues underscore the delicate balance between fostering innovation and protecting creators’ rights as AI continues to reshape the cultural and economic landscape.