AI Wins First Major Court Ruling, But There Is a Catch

A federal judge has delivered a landmark ruling in a closely watched case involving artificial intelligence (AI), declaring that training AI models on copyrighted material can be considered “fair use” under U.S. law. However, the ruling comes with a significant catch: this protection only applies if the developer legally possesses the rights to the content used for training.

In a pre-trial decision in the Northern District of California, U.S. District Judge William Alsup ruled in the case of Bartz v. Anthropic that the AI company’s use of copyrighted books to train its language model, Claude, was “exceedingly transformative” and therefore constituted fair use. The judge’s reasoning was that the AI was not being trained to reproduce the books but to learn from them in order to create new, original content.

What is Fair Use Under US Copyright Law?

Fair use is a legal doctrine that permits the limited use of copyrighted material without permission from the rights holders. It is not an automatic right but is determined by courts on a case-by-case basis by balancing four key factors: 

1. Purpose of the use, including whether it is for commercial or non-profit educational purposes.
2. Nature of the copyrighted work.
3. Amount of the work used.
4. The effect of the use on the potential market for the original work. 

Common examples where fair use may apply include criticism, commentary, news reporting, teaching, scholarship, search engine results, and parody.

The Case for Copyright Infringement

The case, however, is not a complete victory for AI developers. Judge Alsup drew a firm line against the use of pirated materials, ruling that companies could be liable for copyright infringement if they used materials downloaded from illicit “shadow libraries.” The company’s argument that the use of these pirated works was also fair use was rejected, and a trial will proceed to determine damages for this infringement.

This ruling is one of the first of its kind to substantively address the legality of training AI on copyrighted works, a practice that has become standard in the tech industry. The decision provides a potential legal roadmap for AI companies, suggesting that they can mitigate copyright risks by ensuring they have lawfully acquired the data they use for training their models.

The case against Anthropic was brought by a group of authors who alleged that the company’s use of their books without permission constituted copyright infringement. While the fair use ruling on legally acquired content is a setback for the authors, the decision on pirated material offers them a path to hold AI companies accountable.

Long-Term Impact on AI Training

The long-term impact of this ruling is yet to be seen, as it is likely to be appealed. Nevertheless, it sets a significant precedent for the numerous other copyright lawsuits currently pending against AI developers. The decision suggests that while the act of an AI “learning” from copyrighted work may be protected, the origin and legality of the training data itself will be a critical factor in future legal battles.