AI and Copyright Law

AI and copyright law concerns the application of copyright rules to artificial intelligence, covering three connected questions: whether the use of copyrighted works to train AI models infringes copyright, whether AI-generated outputs can be protected by copyright, and who is liable when AI systems reproduce copyrighted material.^[1] The issue has generated more than forty pending lawsuits in the United States alone and sustained policy work by national copyright offices, including a multi-part study by the U.S. Copyright Office.^[2][3]

History and Background

Modern copyright statutes were written with human authors in mind, and the arrival of generative AI — systems trained on massive collections of digitised works — created legal uncertainty on a global scale. In early 2023 the U.S. Copyright Office launched a formal initiative on copyright and AI, issued a notice of inquiry in August 2023, and received more than 10,000 public comments by December 2023.^[1]

The resulting report, Copyright and Artificial Intelligence, is being issued in parts: Part 1 (July 2024) addressed digital replicas; Part 2 (January 2025) addressed the copyrightability of AI-generated outputs; and Part 3 (May 2025) examined generative AI training and fair use.^[1] The Office concluded that "some uses of copyrighted works for generative AI training will qualify as fair use, and some will not", and recommended that licensing markets be allowed to develop without government intervention.^[3]

Key Concepts

The central legal concept in AI training disputes is fair use, which balances the right of reproduction against transformative, socially valuable uses. The Copyright Office found that the steps required to produce a training dataset containing copyrighted works "clearly implicate the right of reproduction", and that whether the copying is defensible depends on factors including transformativeness, commerciality, and the risk of market substitution.^[2][4]

In June 2025, the U.S. District Court for the Northern District of California issued summary judgment in Bartz v. Anthropic PBC, holding that copying books to train the Claude model was fair use because generative AI training is "quintessentially transformative", copying entire books was "reasonably necessary", and the model's outputs did not displace demand for the books.^[3]

A second question is authorship of AI-generated works. The Copyright Office's Part 2 report analysed whether outputs created using generative AI can be protected, concluding that human creative contribution remains essential — a position consistent with most national laws, which require a human author.^[1]

Applications and Impact

The practical impact of AI copyright disputes spans model developers, publishers, visual artists, record labels, and news organisations. More than 40 cases were pending in the United States in mid-2025, with related litigation in at least eight other countries.^[2] In response, many AI companies have moved toward licensed training data, and some models are now trained exclusively on licensed or public-domain works.^[3] Licensing markets, collective management organisations, and industry deals between rightsholders and AI developers have emerged as the principal route for resolving the tension between AI development and copyright protection.^[3][4]

>See Also

• EU AI Act
• AI Watermarking
• PDPA AI Compliance
• Deepfake
• AI Ethics

References